Messaging in Jupyter#

This document explains the basic communications design and messaging specification for how Jupyter frontends and kernels communicate. The ZeroMQ library provides the low-level transport layer over which these messages are sent.


This document contains the authoritative description of the IPython messaging protocol. All developers are strongly encouraged to keep it updated as the implementation evolves, so that we have a single common reference for all protocol details.


The Jupyter message specification is versioned independently of the packages that use it. The current version of the specification is 5.4.


New in and Changed in messages in this document refer to versions of the Jupyter message specification, not versions of jupyter_client.


The basic design is explained in the following diagram:

IPython kernel/frontend messaging architecture.

A single kernel can be simultaneously connected to one or more frontends. The kernel has dedicated sockets for the following functions:

  1. Shell: this single ROUTER socket allows multiple incoming connections from frontends, and this is the socket where requests for code execution, object information, prompts, etc. are made to the kernel by any frontend. The communication on this socket is a sequence of request/reply actions from each frontend and the kernel.

  2. IOPub: this socket is the ‘broadcast channel’ where the kernel publishes all side effects (stdout, stderr, debugging events etc.) as well as the requests coming from any client over the shell socket and its own requests on the stdin socket. There are a number of actions in Python which generate side effects: print() writes to sys.stdout, errors generate tracebacks, etc. Additionally, in a multi-client scenario, we want all frontends to be able to know what each other has sent to the kernel (this can be useful in collaborative scenarios, for example). This socket allows both side effects and the information about communications taking place with one client over the shell channel to be made available to all clients in a uniform manner.

  3. stdin: this ROUTER socket is connected to all frontends, and it allows the kernel to request input from the active frontend when raw_input() is called. The frontend that executed the code has a DEALER socket that acts as a ‘virtual keyboard’ for the kernel while this communication is happening (illustrated in the figure by the black outline around the central keyboard). In practice, frontends may display such kernel requests using a special input widget or otherwise indicating that the user is to type input for the kernel instead of normal commands in the frontend.

    All messages are tagged with enough information (details below) for clients to know which messages come from their own interaction with the kernel and which ones are from other clients, so they can display each type appropriately.

  4. Control: This channel is identical to Shell, but operates on a separate socket to avoid queueing behind execution requests. The control channel is used for shutdown and restart messages, as well as for debugging messages.

    For a smoother user experience, we recommend running the control channel in a separate thread from the shell channel, so that e.g. shutdown or debug messages can be processed immediately without waiting for a long-running shell message to be finished processing (such as an expensive execute request).

  5. Heartbeat: This socket allows for simple bytestring messages to be sent between the frontend and the kernel to ensure that they are still connected.

The actual format of the messages allowed on each of these channels is specified below. Messages are dicts of dicts with string keys and values that are reasonably representable in JSON.

General Message Format#

A message is composed of five dictionaries.

Message Header#

The message header contains information about the message, such as unique identifiers for the originating session and the actual message id, the type of message, the version of the Jupyter protocol, and the date the message was created. In addition, there is a username field, e.g. for the process that generated the message, if applicable. This can be useful in collaborative settings where multiple users may be interacting with the same kernel simultaneously, so that frontends can label the various messages in a meaningful way.

    "msg_id": str,  # typically UUID, must be unique per message
    "session": str,  # typically UUID, should be unique per session
    "username": str,
    # ISO 8601 timestamp for when the message is created
    "date": str,
    # All recognized message type strings are listed below.
    "msg_type": str,
    # the message protocol version
    "version": "5.0",


The session id in a message header identifies a unique entity with state, such as a kernel process or client process.

A client session id, in message headers from a client, should be unique among all clients connected to a kernel. When a client reconnects to a kernel, it should use the same client session id in its message headers. When a client restarts, it should generate a new client session id.

A kernel session id, in message headers from a kernel, should identify a particular kernel process. If a kernel is restarted, the kernel session id should be regenerated.

The session id in a message header can be used to identify the sending entity. For example, if a client disconnects and reconnects to a kernel, and messages from the kernel have a different kernel session id than prior to the disconnect, the client should assume that the kernel was restarted.

Changed in version 5.0: version key added to the header.

Changed in version 5.1: date in the header was accidentally omitted from the spec prior to 5.1, but it has always been in the canonical implementation, so implementers are strongly encouraged to include it. It will be mandatory in 5.1.

Parent header#

When a message is the “result” of another message, such as a side-effect (output or status) or direct reply, the parent_header is a copy of the header of the message that “caused” the current message. _reply messages MUST have a parent_header, and side-effects typically have a parent. If there is no parent, an empty dict should be used. This parent is used by clients to route message handling to the right place, such as outputs to a cell.

    # parent_header is a copy of the request's header
    'msg_id': '...',


The metadata dict contains information about the message that is not part of the content. This is not often used, but can be an extra location to store information about requests and replies, such as extensions adding information about request or execution context.


The content dict is the body of the message. Its structure is dictated by the msg_type field in the header, described in detail for each message below.


Finally, a list of additional binary buffers can be associated with a message. While this is part of the protocol, no official messages make use of these buffers. They are used by extension messages, such as IPython Parallel’s apply and some of ipywidgets’ comm messages.

A full message#

Combining all of these together, a complete message can be represented as the following dictionary of dictionaries (and one list):

    "header" : {
        "msg_id": "...",
        "msg_type": "...",
    "parent_header": {},
    "metadata": {},
    "content": {},
    "buffers": [],


This dictionary structure is not part of the Jupyter protocol that must be implemented by kernels and frontends; that would be The Wire Protocol, which dictates how this information is serialized over the wire. Deserialization is up to the Kernel or frontend implementation, but a dict like this would be a logical choice in most contexts.


Kernels must implement the execute and kernel info messages, along with the associated busy and idle Kernel status messages. All other message types are optional, although we recommend implementing completion if possible. Kernels do not need to send any reply for messages they don’t handle, and frontends should provide sensible behaviour if no reply arrives (except for the required execution and kernel info messages).

stdin messages are unique in that the request comes from the kernel, and the reply from the frontend. The frontend is not required to support this, but if it does not, it must set 'allow_stdin' : False in its execute requests. In this case, the kernel may not send stdin requests. If that field is true, the kernel may send stdin requests and block waiting for a reply, so the frontend must answer.

Both sides should allow unexpected message types, and extra fields in known message types, so that additions to the protocol do not break existing code.

The Wire Protocol#

The above message format is only a logical representation of the contents of Jupyter messages, but does not describe the actual implementation at the wire level in zeromq. This section describes the protocol that must be implemented by Jupyter kernels and clients talking to each other over zeromq.

The reference implementation of the message spec is our Session class.


This section should only be relevant to non-Python consumers of the protocol. Python consumers should import and the use implementation of the wire protocol in jupyter_client.session.Session.

Every message is serialized to a sequence of at least six blobs of bytes:

    b"u-u-i-d",  # zmq identity(ies)
    b"<IDS|MSG>",  # delimiter
    b"baddad42",  # HMAC signature
    b"{header}",  # serialized header dict
    b"{parent_header}",  # serialized parent header dict
    b"{metadata}",  # serialized metadata dict
    b"{content}",  # serialized content dict
    b"\xf0\x9f\x90\xb1"  # extra raw data buffer(s)
    # ...

The front of the message is the ZeroMQ routing prefix, which can be zero or more socket identities. This is every piece of the message prior to the delimiter key <IDS|MSG>. In the case of IOPub, there should be just one prefix component, which is the topic for IOPub subscribers, e.g. execute_result, display_data.


In most cases, the IOPub topics are irrelevant and completely ignored, because frontends just subscribe to all topics. The convention used in the IPython kernel is to use the msg_type as the topic, and possibly extra information about the message, e.g. kernel.{u-u-i-d}.execute_result or stream.stdout

After the delimiter is the HMAC signature of the message, used for authentication. If authentication is disabled, this should be an empty string. By default, the hashing function used for computing these signatures is sha256.


To disable authentication and signature checking, set the key field of a connection file to an empty string.

The signature is the HMAC hex digest of the concatenation of:

  • A shared key (typically the key field of a connection file)

  • The serialized header dict

  • The serialized parent header dict

  • The serialized metadata dict

  • The serialized content dict

In Python, this is implemented via:

# once:
digester = HMAC(key, digestmod=hashlib.sha256)

# for each message
d = digester.copy()
for serialized_dict in (header, parent, metadata, content):
signature = d.hexdigest()

After the signature is the actual message, always in four frames of bytes. The four dictionaries that compose a message are serialized separately, in the order of header, parent header, metadata, and content. These can be serialized by any function that turns a dict into bytes. The default and most common serialization is JSON, but msgpack and pickle are common alternatives.

After the serialized dicts are zero to many raw data buffers, which can be used by message types that support binary data, which can be used in custom messages, such as comms and extensions to the protocol.

Python API#

As messages can be represented as dicts, they map naturally to a func(**kw) call form. We should develop, at a few key points, functional forms of all the requests that take arguments in this manner and automatically construct the necessary dict for sending.

In addition, the Python implementation of the message specification extends messages upon deserialization to the following form for convenience:

  'header' : dict,
  # The msg's unique identifier and type are always stored in the header,
  # but the Python implementation copies them to the top level.
  'msg_id' : str,
  'msg_type' : str,
  'parent_header' : dict,
  'content' : dict,
  'metadata' : dict,
  'buffers': list,

All messages sent to or received by any IPython message handler should have this extended structure.

Messages on the shell (ROUTER/DEALER) channel#


In general, the ROUTER/DEALER sockets follow a request-reply pattern:

The client sends an <action>_request message (such as execute_request) on its shell (DEALER) socket. The kernel receives that request and immediately publishes a status: busy message on IOPub. The kernel then processes the request and sends the appropriate <action>_reply message, such as execute_reply. After processing the request and publishing associated IOPub messages, if any, the kernel publishes a status: idle message. This idle status message indicates that IOPub messages associated with a given request have all been received.

All reply messages have a 'status' field, which will have one of the following values:

  • status='ok': The request was processed successfully, and the remaining content of the reply is specified in the appropriate section below.

  • status='error': The request failed due to an error.

    When status is ‘error’, the usual content of a successful reply should be omitted, instead the following fields should be present:

       'status' : 'error',
       'ename' : str,   # Exception name, as a string
       'evalue' : str,  # Exception value, as a string
       'traceback' : list(str), # traceback frames as strings
  • status='abort': This is the same as status='error' but with no information about the error. No fields should be present other that status.

As a special case, execute_reply messages (see Execution results) have an execution_count field regardless of their status.

Changed in version 5.1: status='abort' has not proved useful, and is considered deprecated. Kernels should send status='error' instead.


This message type is used by frontends to ask the kernel to execute code on behalf of the user, in a namespace reserved to the user’s variables (and thus separate from the kernel’s own internal code and variables).

Message type: execute_request:

content = {
    # Source code to be executed by the kernel, one or more lines.
'code' : str,

# A boolean flag which, if True, signals the kernel to execute
# this code as quietly as possible.
# silent=True forces store_history to be False,
# and will *not*:
#   - broadcast output on the IOPUB channel
#   - have an execute_result
# The default is False.
'silent' : bool,

# A boolean flag which, if True, signals the kernel to populate history
# The default is True if silent is False.  If silent is True, store_history
# is forced to be False.
'store_history' : bool,

# A dict mapping names to expressions to be evaluated in the
# user's dict. The rich display-data representation of each will be evaluated after execution.
# See the display_data content for the structure of the representation data.
'user_expressions' : dict,

# Some frontends do not support stdin requests.
# If this is true, code running in the kernel can prompt the user for input
# with an input_request message (see below). If it is false, the kernel
# should not send these messages.
'allow_stdin' : True,

# A boolean flag, which, if True, aborts the execution queue if an exception is encountered.
# If False, queued execute_requests will execute even if this request generates an exception.
'stop_on_error' : True,

Changed in version 5.0: user_variables removed, because it is redundant with user_expressions.

The code field contains a single string (possibly multiline) to be executed.

The user_expressions field deserves a detailed explanation. In the past, IPython had the notion of a prompt string that allowed arbitrary code to be evaluated, and this was put to good use by many in creating prompts that displayed system status, path information, and even more esoteric uses like remote instrument status acquired over the network. But now that IPython has a clean separation between the kernel and the clients, the kernel has no prompt knowledge; prompts are a frontend feature, and it should be even possible for different frontends to display different prompts while interacting with the same kernel. user_expressions can be used to retrieve this information.

Any error in evaluating any expression in user_expressions will result in only that key containing a standard error message, of the form:

    'status' : 'error',
    'ename' : 'NameError',
    'evalue' : 'foo',
    'traceback' : ...


In order to obtain the current execution counter for the purposes of displaying input prompts, frontends may make an execution request with an empty code string and silent=True.

Upon completion of the execution request, the kernel always sends a reply, with a status code indicating what happened and additional data depending on the outcome. See below for the possible return codes and associated data.

Execution counter (prompt number)#

The kernel should have a single, monotonically increasing counter of all execution requests that are made with store_history=True. This counter is used to populate the In[n] and Out[n] prompts. The value of this counter will be returned as the execution_count field of all execute_reply and execute_input messages.

Execution results#

Message type: execute_reply:

content = {
  # One of: 'ok' OR 'error' OR 'aborted'
  'status' : str,

  # The global kernel counter that increases by one with each request that
  # stores history.  This will typically be used by clients to display
  # prompt numbers to the user.  If the request did not store history, this will
  # be the current value of the counter in the kernel.
  'execution_count' : int,

When status is ‘ok’, the following extra fields are present:

  # 'payload' will be a list of payload dicts, and is optional.
  # payloads are considered deprecated.
  # The only requirement of each payload dict is that it have a 'source' key,
  # which is a string classifying the payload (e.g. 'page').

  'payload' : list(dict),

  # Results for the user_expressions.
  'user_expressions' : dict,

Changed in version 5.0: user_variables is removed, use user_expressions instead.

Payloads (DEPRECATED)#

Execution payloads

Payloads are considered deprecated, though their replacement is not yet implemented.

Payloads are a way to trigger frontend actions from the kernel. Current payloads:

page: display data in a pager.

Pager output is used for introspection, or other displayed information that’s not considered output. Pager payloads are generally displayed in a separate pane, that can be viewed alongside code, and are not included in notebook documents.

    "source": "page",
    # mime-bundle of data to display in the pager.
    # Must include text/plain.
    "data": mimebundle,
    # line offset to start from
    "start": int,

set_next_input: create a new output

used to create new cells in the notebook, or set the next input in a console interface. The main example being %load.

    "source": "set_next_input",
    # the text contents of the cell to create
    "text": "some cell content",
    # If true, replace the current cell in document UIs instead of inserting
    # a cell. Ignored in console UIs.
    "replace": bool,

edit_magic: open a file for editing.

Triggered by %edit. Only the QtConsole currently supports edit payloads.

    "source": "edit_magic",
    "filename": "/path/to/",  # the file to edit
    "line_number": int,  # the line number to start with

ask_exit: instruct the frontend to prompt the user for exit

Allows the kernel to request exit, e.g. via %exit in IPython. Only for console frontends.

    "source": "ask_exit",
    # whether the kernel should be left running, only closing the client
    "keepkernel": bool,


Code can be inspected to show useful information to the user. It is up to the Kernel to decide what information should be displayed, and its formatting.

Message type: inspect_request:

content = {
    # The code context in which introspection is requested
    # this may be up to an entire multiline cell.
    'code' : str,

    # The cursor position within 'code' (in unicode characters) where inspection is requested
    'cursor_pos' : int,

    # The level of detail desired.  In IPython, the default (0) is equivalent to typing
    # 'x?' at the prompt, 1 is equivalent to 'x??'.
    # The difference is up to kernels, but in IPython level 1 includes the source code
    # if available.
    'detail_level' : 0 or 1,

Changed in version 5.0: object_info_request renamed to inspect_request.

Changed in version 5.0: name key replaced with code and cursor_pos, moving the lexing responsibility to the kernel.

Changed in version 5.2: Due to a widespread bug in many frontends, cursor_pos in versions prior to 5.2 is ambiguous in the presence of “astral-plane” characters. In 5.2, cursor_pos must be the actual encoding-independent offset in unicode codepoints. See cursor_pos and unicode offsets for more.

The reply is a mime-bundle, like a display_data message, which should be a formatted representation of information about the context. In the notebook, this is used to show tooltips over function calls, etc.

Message type: inspect_reply:

content = {
    # 'ok' if the request succeeded or 'error', with error information as in all other replies.
    'status' : 'ok',

    # found should be true if an object was found, false otherwise
    'found' : bool,

    # data can be empty if nothing is found
    'data' : dict,
    'metadata' : dict,

Changed in version 5.0: object_info_reply renamed to inspect_reply.

Changed in version 5.0: Reply is changed from structured data to a mime bundle, allowing formatting decisions to be made by the kernel.


Message type: complete_request:

content = {
    # The code context in which completion is requested
    # this may be up to an entire multiline cell, such as
    # 'foo = a.isal'
    'code' : str,

    # The cursor position within 'code' (in unicode characters) where completion is requested
    'cursor_pos' : int,

Changed in version 5.0: line, block, and text keys are removed in favor of a single code for context. Lexing is up to the kernel.

Changed in version 5.2: Due to a widespread bug in many frontends, cursor_pos in versions prior to 5.2 is ambiguous in the presence of “astral-plane” characters. In 5.2, cursor_pos must be the actual encoding-independent offset in unicode codepoints. See cursor_pos and unicode offsets for more.

Message type: complete_reply:

content = {
# status should be 'ok' unless an exception was raised during the request,
# in which case it should be 'error', along with the usual error message content
# in other messages.
'status' : 'ok'

# The list of all matches to the completion request, such as
# ['a.isalnum', 'a.isalpha'] for the above example.
'matches' : list,

# The range of text that should be replaced by the above matches when a completion is accepted.
# typically cursor_end is the same as cursor_pos in the request.
'cursor_start' : int,
'cursor_end' : int,

# Information that frontend plugins might use for extra display information about completions.
'metadata' : dict,

Changed in version 5.0:

  • matched_text is removed in favor of cursor_start and cursor_end.

  • metadata is added for extended information.


For clients to explicitly request history from a kernel. The kernel has all the actual execution history stored in a single location, so clients can request it from the kernel when needed.

Message type: history_request:

content = {

  # If True, also return output history in the resulting dict.
  'output' : bool,

  # If True, return the raw input history, else the transformed input.
  'raw' : bool,

  # So far, this can be 'range', 'tail' or 'search'.
  'hist_access_type' : str,

  # If hist_access_type is 'range', get a range of input cells. session
  # is a number counting up each time the kernel starts; you can give
  # a positive session number, or a negative number to count back from
  # the current session.
  'session' : int,
  # start and stop are line (cell) numbers within that session.
  'start' : int,
  'stop' : int,

  # If hist_access_type is 'tail' or 'search', get the last n cells.
  'n' : int,

  # If hist_access_type is 'search', get cells matching the specified glob
  # pattern (with * and ? as wildcards).
  'pattern' : str,

  # If hist_access_type is 'search' and unique is true, do not
  # include duplicated history.  Default is false.
  'unique' : bool,


New in version 4.0: The key unique for history_request.

Message type: history_reply:

content = {
  # 'ok' if the request succeeded or 'error', with error information as in all other replies.
  'status' : 'ok',

  # A list of 3 tuples, either:
  # (session, line_number, input) or
  # (session, line_number, (input, output)),
  # depending on whether output was False or True, respectively.
  'history' : list,


Most of the history messaging options are not used by Jupyter frontends, and many kernels do not implement them. If you’re implementing these messages in a kernel, the ‘tail’ request is the most useful; this is used by the Qt console, for example. The notebook interface does not use history messages at all.

This interface was designed by exposing all the main options of IPython’s history interface. We may remove some options in a future version of the message spec.

Code completeness#

New in version 5.0.

When the user enters a line in a console style interface, the console must decide whether to immediately execute the current code, or whether to show a continuation prompt for further input. For instance, in Python a = 5 would be executed immediately, while for i in range(5): would expect further input.

There are four possible replies:

  • complete code is ready to be executed

  • incomplete code should prompt for another line

  • invalid code will typically be sent for execution, so that the user sees the error soonest.

  • unknown - if the kernel is not able to determine this. The frontend should also handle the kernel not replying promptly. It may default to sending the code for execution, or it may implement simple fallback heuristics for whether to execute the code (e.g. execute after a blank line).

Frontends may have ways to override this, forcing the code to be sent for execution or forcing a continuation prompt.

Message type: is_complete_request:

content = {
    # The code entered so far as a multiline string
    'code' : str,

Message type: is_complete_reply:

content = {
    # One of 'complete', 'incomplete', 'invalid', 'unknown'
    'status' : str,

    # If status is 'incomplete', indent should contain the characters to use
    # to indent the next line. This is only a hint: frontends may ignore it
    # and use their own autoindentation rules. For other statuses, this
    # field does not exist.
    'indent': str,


Deprecated since version 5.1: connect_request/reply have not proved useful, and are considered deprecated. Kernels are not expected to implement handlers for this message.

When a client connects to the request/reply socket of the kernel, it can issue a connect request to get basic information about the kernel, such as the ports the other ZeroMQ sockets are listening on. This allows clients to only have to know about a single port (the shell channel) to connect to a kernel. The ports for any additional channels the kernel is listening on should be included in the reply. If any ports are omitted from the reply, this indicates that the channels are not running.

Message type: connect_request:

content = {}

For example, a kernel with all channels running:

Message type: connect_reply:

content = {
    'shell_port' : int,   # The port the shell ROUTER socket is listening on.
    'iopub_port' : int,   # The port the PUB socket is listening on.
    'stdin_port' : int,   # The port the stdin ROUTER socket is listening on.
    'hb_port' : int,      # The port the heartbeat socket is listening on.
    'control_port' : int,      # The port the control ROUTER socket is listening on.

Comm info#

When a client needs the currently open comms in the kernel, it can issue a request for the currently open comms. When the optional target_name is specified, the reply only contains the currently open comms for the target.

Message type: comm_info_request:

content = {
    # Optional, the target name
    'target_name': str,

Message type: comm_info_reply:

content = {
    # 'ok' if the request succeeded or 'error', with error information as in all other replies.
    'status' : 'ok',

    # A dictionary of the comms, indexed by uuids.
    'comms': {
        comm_id: {
            'target_name': str,

New in version 5.1.

Kernel info#

If a client needs to know information about the kernel, it can make a request of the kernel’s information. This message can be used to fetch core information of the kernel, including language (e.g., Python), language version number and IPython version number, and the IPython message spec version number.

Message type: kernel_info_request:

content = {

Message type: kernel_info_reply:

content = {
    # 'ok' if the request succeeded or 'error', with error information as in all other replies.
    'status' : 'ok',

    # Version of messaging protocol.
    # The first integer indicates major version.  It is incremented when
    # there is any backward incompatible change.
    # The second integer indicates minor version.  It is incremented when
    # there is any backward compatible change.
    'protocol_version': 'X.Y.Z',

    # The kernel implementation name
    # (e.g. 'ipython' for the IPython kernel)
    'implementation': str,

    # Implementation version number.
    # The version number of the kernel's implementation
    # (e.g. IPython.__version__ for the IPython kernel)
    'implementation_version': 'X.Y.Z',

    # Information about the language of code for the kernel
    'language_info': {
        # Name of the programming language that the kernel implements.
        # Kernel included in IPython returns 'python'.
        'name': str,

        # Language version number.
        # It is Python version number (e.g., '2.7.3') for the kernel
        # included in IPython.
        'version': 'X.Y.Z',

        # mimetype for script files in this language
        'mimetype': str,

        # Extension including the dot, e.g. '.py'
        'file_extension': str,

        # Pygments lexer, for highlighting
        # Only needed if it differs from the 'name' field.
        'pygments_lexer': str,

        # Codemirror mode, for highlighting in the notebook.
        # Only needed if it differs from the 'name' field.
        'codemirror_mode': str or dict,

        # Nbconvert exporter, if notebooks written with this kernel should
        # be exported with something other than the general 'script'
        # exporter.
        'nbconvert_exporter': str,

    # A banner of information about the kernel,
    # which may be displayed in console environments.
    'banner': str,

    # A boolean flag which tells if the kernel supports debugging in the notebook.
    # Default is False
    'debugger': bool,

    # Optional: A list of dictionaries, each with keys 'text' and 'url'.
    # These will be displayed in the help menu in the notebook UI.
    'help_links': [
        {'text': str, 'url': str}

Refer to the lists of available Pygments lexers and codemirror modes for those fields.

Changed in version 5.0: Versions changed from lists of integers to strings.

Changed in version 5.0: ipython_version is removed.

Changed in version 5.0: language_info, implementation, implementation_version, banner and help_links keys are added.

Changed in version 5.0: language_version moved to language_info.version

Changed in version 5.0: language moved to

Messages on the Control (ROUTER/DEALER) channel#

Kernel shutdown#

The clients can request the kernel to shut itself down; this is used in multiple cases:

  • when the user chooses to close the client application via a menu or window control.

  • when the user types ‘exit’ or ‘quit’ (or their uppercase magic equivalents).

  • when the user chooses a GUI method (like the ‘Ctrl-C’ shortcut in the IPythonQt client) to force a kernel restart to get a clean kernel without losing client-side state like history or inlined figures.

Implementation recommendation for starting kernels: A restart should optimally preserve as many resources outside the kernel as possible (e.g. only restart the kernel and its subprocesses and not any parent processes). That is, ideally a restart should be “in-place”. For local kernels, there is typically no parent process so a “hard” restart and an in-place restart are identical whereas for remote kernels this is not generally the same. As an example, if a remote kernel is run in a container, during an in-place restart the container may be kept running and a new kernel process within it would be started.

The client sends a shutdown request to the kernel, and once it receives the reply message (which is otherwise empty), it can assume that the kernel has completed shutdown safely. The request is sent on the control channel.

Upon their own shutdown, client applications will typically execute a last minute sanity check and forcefully terminate any kernel that is still alive, to avoid leaving stray processes in the user’s machine.

Message type: shutdown_request:

content = {
    'restart' : bool # False if final shutdown, or True if shutdown precedes a restart

Message type: shutdown_reply:

content = {
    # 'ok' if the request succeeded or 'error', with error information as in all other replies.
    'status' : 'ok',

    'restart' : bool # False if final shutdown, or True if shutdown precedes a restart


When the clients detect a dead kernel thanks to inactivity on the heartbeat socket, they simply send a forceful process termination signal, since a dead process is unlikely to respond in any useful way to messages.

Changed in version 5.4: Sending a shutdown_request message on the shell channel is deprecated.

Kernel interrupt#

In case a kernel can not catch operating system interrupt signals (e.g. the used runtime handles signals and does not allow a user program to define a callback), a kernel can choose to be notified using a message instead. For this to work, the kernels kernelspec must set interrupt_mode to message. An interruption will then result in the following message on the control channel:

Message type: interrupt_request:

content = {}

Message type: interrupt_reply:

content = {
    # 'ok' if the request succeeded or 'error', with error information as in all other replies.
    'status' : 'ok'

New in version 5.3.

Debug request#

This message type is used with debugging kernels to request specific actions to be performed by the debugger such as adding a breakpoint or stepping into a code.

Message type: debug_request:

content = {}

Message type: debug_reply:

content = {}

The content dicts of the debug_request and debug_reply messages respectively follow the specification of the Request and Response messages from the Debug Adapter Protocol (DAP) as of version 1.39 or later.

Debug requests and replies are sent over the control channel to prevent queuing behind execution requests.

Additions to the DAP#

The Jupyter debugger protocol makes several additions to the DAP:


In order to support the debugging of notebook cells and of Jupyter consoles, which are not based on source files, we need a message to submit code to the debugger to which breakpoints can be added.

Content of the dumpCell request:

    'type' : 'request',
    'command' : 'dumpCell',
    'arguments' : {
        'code' : str  # the content of the cell being submitted.

Content of the dumpCell response:

     'type' : 'response',
     'success': bool,
     'body': {
         'sourcePath': str  # filename for the dumped source

In order to support page reloading, or a client connecting at a later stage, Jupyter kernels must store the state of the debugger (such as breakpoints, whether the debugger is currently stopped). The debugInfo request is a DAP Request with no extra argument.

Content of the debugInfo request:

    'type' : 'request',
    'command' : 'debugInfo'

Content of the debugInfo response:

    'type' : 'response',
    'success' : bool,
    'body' : {
        'isStarted' : bool,  # whether the debugger is started,
        'hashMethod' : str,  # the hash method for code cell. Default is 'Murmur2',
        'hashSeed' : str,  # the seed for the hashing of code cells,
        'tmpFilePrefix' : str,  # prefix for temporary file names
        'tmpFileSuffix' : str,  # suffix for temporary file names
        'breakpoints' : [  # breakpoints currently registered in the debugger.
                'source' : str,  # source file
                'breakpoints' : list(source_breakpoints)  # list of breakpoints for that source file
        'stoppedThreads' : list(int),  # threads in which the debugger is currently in a stopped state
        'richRendering' : bool,  # whether the debugger supports rich rendering of variables
        'exceptionPaths' : list(str),  # exception names used to match leaves or nodes in a tree of exception

The source_breakpoint schema is specified by the Debug Adapter Protocol.


The inspectVariables is meant to retrieve the values of all the variables that have been defined in the kernel. It is a DAP Request with no extra argument.

Content of the inspectVariables request:

    'type' : 'request',
    'command' : 'inspectVariables'

Content of the inspectVariables response:

    'type' : 'response',
    'success' : bool,
    'body' : {
        'variables' : [ # variables defined in the notebook.
                'name' : str,
                'variablesReference' : int,
                'value' : str,
                'type' : str

The richInspectVariables request allows to get the rich representation of a variable that has been defined in the kernel.

Content of the richInspectVariables request:

    'type' : 'request',
    'command' : 'richInspectVariables',
    'arguments' : {
        'variableName' : str,
        # The frameId is used when the debugger hit a breakpoint only.
        'frameId' : int

Content of the richInspectVariables response:

    'type' : 'response',
    'success' : bool,
    'body' : {
        # Dictionary of rich representations of the variable
        'data' : dict,
        'metadata' : dict

The copyToGlobals request allows to copy a variable from the local variable panel of the debugger to the global scope to inspect it after debug session.

Content of the copyToGlobals request:

    'type': 'request',
    'command': 'copyToGlobals',
    'arguments': {
        # the variable to copy from the frame corresponding to `srcFrameId`
        'srcVariableName': str,
        'srcFrameId': int,
        # the copied variable name in the global scope
        'dstVariableName': str

Content of the copyToGlobals response:

    'type': 'response',
    'success': bool,
    'command': 'setExpression',
    'body': {
        # string representation of the copied variable
        'value': str,
        # type of the copied variable
        'type': str,
        'variablesReference': int

New in version 5.5.

Messages on the IOPub (PUB/SUB) channel#

Streams (stdout, stderr, etc)#

Message type: stream:

content = {
    # The name of the stream is one of 'stdout', 'stderr'
    'name' : str,

    # The text is an arbitrary string to be written to that stream
    'text' : str,

Changed in version 5.0: ‘data’ key renamed to ‘text’ for consistency with the notebook format.

Display Data#

This type of message is used to bring back data that should be displayed (text, html, svg, etc.) in the frontends. This data is published to all frontends. Each message can have multiple representations of the data; it is up to the frontend to decide which to use and how. A single message should contain all possible representations of the same information. Each representation should be a JSON’able data structure, and should be a valid MIME type.

Some questions remain about this design:

  • Do we use this message type for execute_result/displayhook? Probably not, because the displayhook also has to handle the Out prompt display. On the other hand we could put that information into the metadata section.

Message type: display_data:

content = {

    # Who create the data
    # Used in V4. Removed in V5.
    # 'source' : str,

    # The data dict contains key/value pairs, where the keys are MIME
    # types and the values are the raw data of the representation in that
    # format.
    'data' : dict,

    # Any metadata that describes the data
    'metadata' : dict,

    # Optional transient data introduced in 5.1. Information not to be
    # persisted to a notebook or other documents. Intended to live only
    # during a live kernel session.
    'transient': dict,

The metadata contains any metadata that describes the output. Global keys are assumed to apply to the output as a whole. The metadata dict can also contain mime-type keys, which will be sub-dictionaries, which are interpreted as applying only to output of that type. Third parties should put any data they write into a single dict with a reasonably unique name to avoid conflicts.

The only metadata keys currently defined in IPython are the width and height of images:

metadata = {
  'image/png' : {
    'width': 640,
    'height': 480

and expanded for JSON data:

metadata = {
  'application/json' : {
    'expanded': True

The transient dict contains runtime metadata that should not be persisted to document formats and is fully optional. The only transient key currently defined in Jupyter is display_id:

transient = {
    'display_id': 'abcd'

Changed in version 5.0: application/json data should be unpacked JSON data, not double-serialized as a JSON string.

Changed in version 5.1: transient is a new field.

Update Display Data#

New in version 5.1.

Displays can now be named with a display_id within the transient field of display_data or execute_result.

When a display_id is specified for a display, it can be updated later with an update_display_data message. This message has the same format as display_data messages and must contain a transient field with a display_id.

Message type: update_display_data:

content = {

    # The data dict contains key/value pairs, where the keys are MIME
    # types and the values are the raw data of the representation in that
    # format.
    'data' : dict,

    # Any metadata that describes the data
    'metadata' : dict,

    # Any information not to be persisted to a notebook or other environment
    # Intended to live only during a kernel session
    'transient': dict,

Frontends can choose how they update prior outputs (or if they regard this as a regular display_data message). Within the jupyter and nteract notebooks, all displays that match the display_id are updated (even if there are multiple).

Code inputs#

To let all frontends know what code is being executed at any given time, these messages contain a re-broadcast of the code portion of an execute_request, along with the execution_count.

Message type: execute_input:

content = {
    'code' : str,  # Source code to be executed, one or more lines

    # The counter for this execution is also provided so that clients can
    # display it, since IPython automatically creates variables called _iN
    # (for input prompt In[N]).
    'execution_count' : int

Changed in version 5.0: pyin is renamed to execute_input.

Execution results#

Results of an execution are published as an execute_result. These are identical to display_data messages, with the addition of an execution_count key.

Results can have multiple simultaneous formats depending on its configuration. A plain text representation should always be provided in the text/plain mime-type. Frontends are free to display any or all of these according to its capabilities. Frontends should ignore mime-types they do not understand. The data itself is any JSON object and depends on the format. It is often, but not always a string.

Message type: execute_result:

content = {

    # The counter for this execution is also provided so that clients can
    # display it, since IPython automatically creates variables called _N
    # (for prompt N).
    'execution_count' : int,

    # data and metadata are identical to a display_data message.
    # the object being displayed is that passed to the display hook,
    # i.e. the *result* of the execution.
    'data' : dict,
    'metadata' : dict,

Execution errors#

When an error occurs during code execution

Message type: error:

content = {
   # Similar content to the execute_reply messages for the 'error' case,
   # except the 'status' and 'execution_count' fields are omitted.

Changed in version 5.0: pyerr renamed to error

Kernel status#

This message type is used by frontends to monitor the status of the kernel.

Message type: status:

content = {
    # When the kernel starts to handle a message, it will enter the 'busy'
    # state and when it finishes, it will enter the 'idle' state.
    # The kernel will publish state 'starting' exactly once at process startup.
    execution_state : ('busy', 'idle', 'starting')

When a kernel receives a request and begins processing it, the kernel shall immediately publish a status message with execution_state: 'busy'. When that kernel has completed processing the request and has finished publishing associated IOPub messages, if any, it shall publish a status message with execution_state: 'idle'. Thus, the outputs associated with a given execution shall generally arrive between the busy and idle status messages associated with a given request.


A caveat for asynchronous output

Asynchronous output (e.g. from background threads) may be produced after the kernel has sent the idle status message that signals the completion of the request. The handling of these out-of-order output messages is currently undefined in this specification, but the Jupyter Notebook continues to handle IOPub messages associated with a given request after the idle message has arrived, as long as the output area corresponding to that request is still active.

Changed in version 5.0: Busy and idle messages should be sent before/after handling every request, not just execution.

Clear output#

This message type is used to clear the output that is visible on the frontend.

Message type: clear_output:

content = {

    # Wait to clear the output until new output is available.  Clears the
    # existing output immediately before the new output is displayed.
    # Useful for creating simple animations with minimal flickering.
    'wait' : bool,

Changed in version 4.1: stdout, stderr, and display boolean keys for selective clearing are removed, and wait is added. The selective clearing keys are ignored in v4 and the default behavior remains the same, so v4 clear_output messages will be safely handled by a v4.1 frontend.

Debug event#

This message type is used by debugging kernels to send debugging events to the frontend.

Message type: debug_event:

content = {}

The content dict follows the specification of the Event message from the Debug Adapter Protocol (DAP).

New in version 5.5.

Messages on the stdin (ROUTER/DEALER) channel#

With the stdin ROUTER/DEALER socket, the request/reply pattern goes in the opposite direction of most kernel communication. With the stdin socket, the kernel makes the request, and the single frontend provides the response. This pattern allows code to prompt the user for a line of input, which would normally be read from stdin in a terminal.

Many programming languages provide a function which displays a prompt, blocks until the user presses return, and returns the text they typed before pressing return. In Python 3, this is the input() function; in R it is called readline(). If the execute_request message has allow_stdin==True, kernels may implement these functions so that they send an input_request message and wait for a corresponding input_reply. The frontend is responsible for displaying the prompt and getting the user’s input.

If allow_stdin is False, the kernel must not send stdin_request. The kernel may decide what to do instead, but it’s most likely that calls to the ‘prompt for input’ function should fail immediately in this case.

Message type: input_request:

content = {
    # the text to show at the prompt
    'prompt' : str,
    # Is the request for a password?
    # If so, the frontend shouldn't echo input.
    'password' : bool

Message type: input_reply:

content = { 'value' : str }

When password is True, the frontend should not show the input as it is entered. Different frontends may obscure it in different ways; e.g. showing each character entered as the same neutral symbol, or not showing anything at all as the user types.

Changed in version 5.0: password key added.


The stdin socket of the client is required to have the same zmq IDENTITY as the client’s shell socket. Because of this, the input_request must be sent with the same IDENTITY routing prefix as the execute_reply in order for the frontend to receive the message.


This pattern of requesting user input is quite different from how stdin works at a lower level. The Jupyter protocol does not support everything code running in a terminal can do with stdin, but we believe that this enables the most common use cases.

Heartbeat for kernels#

Clients send ping messages on a REQ socket, which are echoed right back from the Kernel’s REP socket. These are simple bytestrings, not full JSON messages described above.

Custom Messages#

New in version 4.1.

Message spec 4.1 (IPython 2.0) added a messaging system for developers to add their own objects with Frontend and Kernel-side components, and allow them to communicate with each other. To do this, IPython adds a notion of a Comm, which exists on both sides, and can communicate in either direction.

These messages are fully symmetrical - both the Kernel and the Frontend can send each message, and no messages expect a reply. The Kernel listens for these messages on the Shell channel, and the Frontend listens for them on the IOPub channel.

Opening a Comm#

Opening a Comm produces a comm_open message, to be sent to the other side:

  'comm_id' : 'u-u-i-d',
  'target_name' : 'my_comm',
  'data' : {}

Every Comm has an ID and a target name. The code handling the message on the receiving side is responsible for maintaining a mapping of target_name keys to constructors. After a comm_open message has been sent, there should be a corresponding Comm instance on both sides. The data key is always a dict and can be any extra JSON information used in initialization of the comm.

If the target_name key is not found on the receiving side, then it should immediately reply with a comm_close message to avoid an inconsistent state.

Comm Messages#

Comm messages are one-way communications to update comm state, used for synchronizing widget state, or simply requesting actions of a comm’s counterpart.

Essentially, each comm pair defines their own message specification implemented inside the data dict.

There are no expected replies (of course, one side can send another comm_msg in reply).

Message type: comm_msg:

  'comm_id' : 'u-u-i-d',
  'data' : {}

Tearing Down Comms#

Since comms live on both sides, when a comm is destroyed the other side must be notified. This is done with a comm_close message.

Message type: comm_close:

  'comm_id' : 'u-u-i-d',
  'data' : {}

Output Side Effects#

Since comm messages can execute arbitrary user code, handlers should set the parent header and publish status busy / idle, just like an execute request.


5.5 (draft)#

  • Added debug_request/reply messages

  • Added debug_event message


  • Sending a shutdown_request message on the shell channel is deprecated. It should be sent on the control channel.


  • Kernels can now opt to be interrupted by a message sent on the control channel instead of a system signal. See Kernel specs and Kernel interrupt.


  • Resolve ambiguity of cursor_pos field in the presence of unicode surrogate pairs. In 5.2, cursor_pos must be the actual encoding-independent offset in unicode codepoints.


  • date in the header was accidentally omitted from the spec prior to 5.1, but it has always been in the canonical implementation, so implementers are strongly encouraged to include it. It is mandatory in 5.1.

  • status='abort' in replies has not proved useful, and is considered deprecated. Kernels should send status='error' instead.

  • comm_info_request/reply added

  • connect_request/reply have not proved useful, and are considered deprecated. Kernels are not expected to implement handlers for this message.

  • new transient field in display_data

  • new update_display_data message


General changes:

  • version key added to message headers

  • busy and idle status messages should be sent before/after handling every request, not just execution

Message renames to remove Python-specific-ness:

  • pyin message renamed to execute_input

  • pyerr renamed to error

  • object_info_request/reply messages renamed to inspect_request/reply

Kernel info:

  • versions changed from lists of integers to strings

  • ipython_version is removed

  • language_info, implementation, implementation_version, banner

    and help_links keys are added.

  • language_version is moved to language_info.version

  • language is moved to


  • user_variables is removed from execute_request/reply because it is redundant with user_expressions

  • password key added to input_request


  • data key in stream messages renamed to text for consistency with the notebook format.

  • application/json in mimebundles should be unpacked JSON data, not a double-serialized JSON string.


  • name key in inspect_request replaced with code and cursor_pos, moving the lexing responsibility to the kernel.

  • object_info_reply is now a mimebundle, allowing formatting decisions to be made by the kernel.


  • complete_request: line, block, and text keys are removed in

    favor of a single code for context. Lexing is up to the kernel.

  • complete_reply:
    • matched_text is removed in favor of cursor_start and cursor_end.

    • metadata is added for extended information.

  • new is_complete_request and is_complete_reply messages


  • comm_open/close/msg messages added

  • clear_output: stdout, stderr, and display boolean keys for selective clearing are removed, and wait is added. The selective clearing keys are ignored in v4 and the default behavior remains the same, so v4 clear_output messages will be safely handled by a v4.1 frontend.


cursor_pos and unicode offsets#

Many frontends, especially those implemented in javascript, reported cursor_pos as the interpreter’s string index, which is not the same as the unicode character offset if the interpreter uses UTF-16 (e.g. javascript or Python 2 on macOS), which stores “astral-plane” characters such as 𝐚 (U+1D41A) as surrogate pairs, taking up two indices instead of one, causing a unicode offset drift of one per astral-plane character. Not all frontends have this behavior, however, and after JSON serialization information about which encoding was used when calculating the offset is lost, so assuming cursor_pos is calculated in UTF-16 could result in a similarly incorrect offset for frontends that did the right thing.

For this reason, in protocol versions prior to 5.2, cursor_pos is officially ambiguous in the presence of astral plane unicode characters. Frontends claiming to implement protocol 5.2 MUST identify cursor_pos as the encoding-independent unicode character offset. Kernels may choose to expect the UTF-16 offset from requests implementing protocol 5.1 and earlier, in order to behave correctly with the most popular frontends. But they should know that doing so introduces the inverse bug for the frontends that do not have this bug.

As an example, use a python3 kernel and evaluate 𨭎𨭎𨭎𨭎𨭎 = 10. Then type 𨭎𨭎 followed by the tab key and see if it properly completes.

Known affected frontends (as of 2017-06):

  • Jupyter Notebook < 5.1

  • JupyterLab < 0.24

  • nteract < 0.2.0

  • Jupyter Console and QtConsole with Python 2 on macOS and Windows

Known not affected frontends:

  • QtConsole, Jupyter Console with Python 3 or Python 2 on Linux, CoCalc