Dr. Dominik Jain e20a0d748e
✨ Add OAuth support for MCP server (#12480)
* 🎉 Add OAuth support for remote MCP connections

Add MCP-specific discovery, client registration, and approval endpoints.
Exchange single-use authorization codes with PKCE for the fixed MCP user
token, enabling MCP and creating the token on approval when needed.

Use self-contained client IDs without new database tables or stored client
records. Keep temporary authorization state in Redis/Valkey.

Accept bearer credentials while preserving query-parameter connections.
Enable OAuth through PENPOT_MCP_AUTH_URI in multi-user mode. Document the
environment variable in the README and implementation details in Serena.

Explain protocol and concurrency decisions in the Clojure code. Cover token
provisioning, OAuth state, client registration, HTTP flows, and the approval
page with tests.

AI-assisted-by: gpt-6

* 🐛 Wait for devenv initialization before starting builds

Docker reports a container as Running as soon as its main process starts,
while the entrypoint may still be preparing the development environment.
manage.sh used that status to launch tmux jobs, allowing frontend builds
to race with ownership changes to /opt/rustup and Cargo setup.

On a fresh container, rustup could fail with Permission denied before the
frontend watcher started. Nginx would later serve the existing JavaScript,
leaving source changes absent from the running app despite restarting
devenv.

Wait for nginx's local health endpoint before launching the jobs. Nginx
starts after entrypoint setup, so its response confirms that setup has
finished. Allow up to 120 seconds and report a timeout without starting
the jobs if initialization does not complete.

Validation: shell syntax and whitespace checks pass.

AI-assisted-by: gpt-6

* ♻️ Share the login modal with MCP authorization

Extract the viewer login modal and its styles into a shared auth
component, retaining the viewer wrapper and its close behavior.

Use the modal on the MCP approval page and preserve the pending request
through login. Closing it clears the saved return URL and shows a
cancellation message without further actions.

Cover return URL handling and cancellation with frontend tests.

AI-assisted-by: gpt-6

* ✨ Show client icons in MCP authorization

Preserve HTTPS logo URLs from client registration and published metadata
and show the icon beside the application name on the approval page.
Ignore unusable URLs and hide failed images without blocking approval.

Update the access description to refer to MCP connections established
from Penpot. Cover logo metadata and approval responses in backend tests.

AI-assisted-by: gpt-6

* 🐛 Follow the system theme when no preference is configured

Default to the system theme in the shared resolver used by the app and
plugins. Preserve explicit light and dark choices, including the legacy
default value that means dark.

Reuse the reactive theme hook in progress widgets and show the system
preference in settings for accounts without a configured theme.

Verify preference resolution with regression tests and browser checks.
All 1,474 frontend tests pass.

AI-assisted-by: gpt-6

* 🐛 Replace expired MCP tokens during OAuth approval

Settings can give an MCP token an expiry date. OAuth approval reused
the newest MCP token without checking it, so an expired token reached
the client and never matched a plugin connection.

Approval now treats an expired token as missing and creates a new one.
Token exchange rejects a code whose token expired after approval with
invalid_grant.

AI-assisted-by: claude-opus-5-5

* 🐛 Log unexpected errors on MCP OAuth endpoints

The OAuth routes sit outside the shared HTTP middleware, so the shared
error handler never saw them, and their own handler turned unexpected
failures into a 503 without logging anything. Failures in the session
middleware of the approval endpoints had no handler at all.

Unexpected failures now also go through the shared error handler for
its logging, while clients still get an OAuth error body. The OAuth
error handler also wraps the session middleware of the approval
endpoints.

AI-assisted-by: claude-opus-5-5

* 🔧 Add devenv script to run the MCP server with OAuth

start-mcp-devenv-oauth starts the MCP server in multi-user mode with
PENPOT_MCP_AUTH_URI derived from PENPOT_PUBLIC_URI, to test OAuth in
the devenv. Agentic tooling still uses the single-user
start-mcp-devenv.

AI-assisted-by: claude-opus-5-5

* 🐛 Fix failing test case

---------

Co-authored-by: alonso.torres <alonso.torres@kaleidos.net>
2026-10-28 18:40:04 +01:00
..
2026-02-04 12:22:36 +01:00
2026-04-20 19:37:02 +02:00
2026-02-04 12:22:36 +01:00
2026-02-04 12:22:36 +01:00
2026-10-07 10:41:29 +02:00

mcp-server-cover-github-1

Penpot's Official MCP Server

Penpot integrates a LLM layer built on the Model Context Protocol (MCP) via Penpot's Plugin API to interact with a Penpot design file. Penpot's MCP server enables LLMs to perform data queries, transformation and creation operations.

Penpot's MCP Server is unlike any other you've seen. You get design-to- design, code-to-design and design-code supercharged workflows.

Penpot MCP video playlist

Architecture

The Penpot MCP Server exposes tools to AI clients (LLMs), which support the retrieval of design data as well as the modification and creation of design elements. The MCP server communicates with Penpot via the dedicated Penpot MCP Plugin, which connects to the MCP server via WebSocket.
This enables the LLM to carry out tasks in the context of a design file by executing code that leverages the Penpot Plugin API. The LLM is free to write and execute arbitrary code snippets within the Penpot Plugin environment to accomplish its tasks.

Architecture

This repository thus contains not only the MCP server implementation itself but also the supporting Penpot MCP Plugin (see section Repository Structure below).

Demonstration

Video

Usage

Important

These instructions are for local MCP server usage only!

The instructions below are for users who want to run the server locally, allowing the server to have access to local files, which can be relevant for development/coding tasks.

If you need to work only on designs, the centrally hosted remote MCP server should be considered instead. Please refer to the MCP usage instructions in the help center.

To run the Penpot MCP server locally, you must

  • run the MCP server and connect your AI client to it,
  • run the web server providing the Penpot MCP plugin, and
  • open the Penpot MCP plugin in Penpot and connect it to the MCP server.

Follow the steps below to enable the integration.

Prerequisites

The project requires Node.js 20 or later (tested with v22.x).

1. Starting the MCP Server and the Plugin Server

Running a Released Version via npx

The easiest way to launch the servers is to use npx to run the appropriate version that matches your Penpot version.

If you are using the latest Penpot release, e.g. as served on design.penpot.app, run:

npx -y @penpot/mcp@latest

Once the servers are running, continue with step 2.

Running the Source Version from the Repository

The tools pnpm and npx should be available in your terminal.

On Windows, use the Git Bash terminal to ensure compatibility with the provided scripts.

Clone the Appropriate Branch of the Repository

Clone the Penpot repository, using the proper branch/tag depending on the version of Penpot you want to use the MCP server with.
For instance, to target the latest development version, use the develop branch:

git clone https://github.com/penpot/penpot.git --branch develop --depth 1

Then change into the mcp directory:

cd penpot/mcp
Build & Launch the MCP Server and the Plugin Server

If it's your first execution, install the required dependencies. (If you are using the Penpot devenv, this step is not necessary, as dependencies are already installed.)

./scripts/setup

Then build all components and start the two servers:

pnpm run bootstrap

This bootstrap command will:

  • install dependencies for all components
  • build all components
  • start all components

2. Load the Plugin in Penpot and Establish the Connection

Note

Browser Connectivity Restrictions

Starting with Chromium version 142, the private network access (PNA) restrictions have been hardened, and when connecting to localhost from a web application served from a different origin (such as https://design.penpot.app), the connection must explicitly be allowed.

Most Chromium-based browsers (e.g. Chrome, Vivaldi) will display a popup requesting permission to access the local network. Be sure to approve the request to allow the connection.

Some browsers take additional security measures, and you may need to disable them. For example, in Brave, disable the "Shield" for the Penpot website to allow local network access.

If your browser refuses to connect to the locally served plugin, check its configuration or try a different browser (e.g. Firefox) that does not enforce these restrictions.

  1. Open Penpot in your browser
  2. Navigate to a design file
  3. Open the Plugins menu
  4. Load the plugin using the development URL (http://localhost:4400/manifest.json by default)
  5. Open the plugin UI
  6. In the plugin UI, click "Connect to MCP server". The connection status should change from "Not connected" to "Connected". (Check the browser's developer console for WebSocket connection logs. Check the MCP server terminal for WebSocket connection messages.)

Important

Do not close the plugin's UI while using the MCP server, as this will close the connection. Also keep the Penpot tab active during long MCP sessions. Browsers may freeze, suspend, or unload inactive tabs to save resources; when that happens, the MCP server will reject tasks until the tab wakes up or reconnects. In Chrome, add your Penpot site to Settings → Performance → Always keep these sites active or pin the tab to reduce tab deactivation.

3. Connect an MCP Client

Important

Use an appropriate model.

We recommend that you ...

  • use the most capable model at your disposal. You will achieve the best results with frontier models, especially when dealing with more complex tasks. Weaker models, including most locally hosted ones, are unlikely to produce usable results for anything beyond simple tasks.
  • use a vision language model (VLM), as many design tasks necessitate visual inspection. (If you are using a standard commercial model, it almost certainly supports vision already.)

By default, the server provides a Streamable HTTP endpoint at http://localhost:4401/mcp. Simply configure the client to connect the MCP server by providing the respective URL.

The legacy /sse and /messages endpoints are no longer supported. Clients using the legacy SSE transport must switch to Streamable HTTP at /mcp.

You can change the port by setting the PENPOT_MCP_SERVER_PORT environment variable before starting the server. This endpoint can be used directly by MCP clients that support Streamable HTTP.

Automatically configuring your client

You can configure your client with the add-mcp helper. Simply call

npx -y add-mcp -g -n penpot http://localhost:4401/mcp

and follow the interactive dialogue to configure the clients of your choice. The config entry name is penpot (override it with -n <name>) and the URL points to the local http endpoint (adjust the port if you changed PENPOT_MCP_SERVER_PORT).

When using a client that only supports stdio transport like Claude Desktop, a proxy like mcp-remote is required. More information on connecting your client follows below.

Using a Proxy for stdio Transport

The mcp-remote package can proxy stdio transport to Streamable HTTP, allowing clients that support only stdio to connect to the MCP server indirectly. Use it to provide the launch command for your MCP client as follows:

    npx -y mcp-remote http://localhost:4401/mcp --allow-http

Example: Claude Desktop

For Windows and macOS, there is the official Claude Desktop app, which you can use as an MCP client. For Linux, there is an unofficial community version.

Since Claude Desktop natively supports only stdio transport, you will need to use a proxy like mcp-remote. Install it as described above.

To add the server to Claude Desktop's configuration, locate the configuration file (or find it via Menu / File / Settings / Developer):

  • Windows: %APPDATA%/Claude/claude_desktop_config.json
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Linux: ~/.config/Claude/claude_desktop_config.json

Add a penpot entry under mcpServers with the following content:

{
    "mcpServers": {
        "penpot": {
            "command": "npx",
            "args": ["-y", "mcp-remote", "http://localhost:4401/mcp", "--allow-http"]
        }
    }
}

After updating the configuration file, restart Claude Desktop completely for the changes to take effect.

Important

Be sure to fully quit the app for the changes to take effect; closing the window is not sufficient.
To fully terminate the app, choose Menu / File / Quit.

After the restart, you should see the MCP server listed when clicking on the "Search and tools" icon at the bottom of the prompt input area.

Working with Multiple Files

Connect the files you want to use:

  • Open each file in a separate Penpot tab, run the plugin, and connect it to the same MCP server.
  • Each browser tab uses a short session ID derived only from penpot.currentUser.sessionId, shown in the plugin UI with a Copy button. The ID stays the same across workspace navigation.
  • Multiple tabs can connect to the same file. Connecting or disconnecting one tab does not disconnect the others.

Choose a session for your agent:

  • With exactly one connected session, the server selects it automatically when the agent omits the session ID.
  • With several connected sessions, the server lists them and asks the agent to have you select one.
  • You can also copy an ID from the plugin and include it in your prompt, for example: "Use Penpot session o37vgcqsvt and list the pages in that file."

For integrated remote MCP connection controls and session behavior, see the Help Center guide.

Repository Structure

This repository is a monorepo containing four main components:

  1. Common Types (packages/common/):

    • Shared TypeScript definitions for request/response protocol
    • Ensures type safety across server and plugin components
  2. Penpot MCP Server (packages/server/):

    • Provides MCP tools to LLMs for Penpot interaction
    • Runs a WebSocket server accepting connections from the Penpot MCP plugin
    • Implements request/response correlation with unique task IDs
    • Handles task timeouts and proper error reporting
  3. Penpot MCP Plugin (packages/plugin/):

    • Connects to the MCP server via WebSocket
    • Executes tasks in Penpot using the Plugin API
    • Sends structured responses back to the server#
  4. Types Generator (types-generator/):

    • Generates data on API types for the MCP server (development use)

The core components are written in TypeScript, rendering interactions with the Penpot Plugin API both natural and type-safe.

Configuration

The Penpot MCP server can be configured using environment variables.

Server Configuration

Environment Variable Description Default
PENPOT_MCP_SERVER_HOST Address on which the MCP server listens (binds to) localhost
PENPOT_MCP_SERVER_PORT Port for the MCP server 4401
PENPOT_MCP_WEBSOCKET_PORT Port for the WebSocket server (plugin connection) 4402
PENPOT_MCP_REPL_PORT Port for the REPL server (development/debugging) 4403
PENPOT_MCP_REPL_HOST Address on which the REPL server listens (binds to) localhost
PENPOT_MCP_REPL_ENABLE Explicitly enable/disable the REPL server. Set to true to enable. When unset, defaults to the value of PENPOT_MCP_DEVENV. The REPL server never starts in multi-user mode. (unset)
PENPOT_MCP_REMOTE_MODE Enable remote mode (disables file system access). Set to true to enable. false
PENPOT_MCP_DEVENV Enable Penpot development environment tools in local single-user mode. Set to true to enable. false
PENPOT_MCP_TOOL_TIMEOUT_S Timeout, in seconds, for tool calls dispatched to the Penpot plugin 120
PENPOT_MCP_EXPORT_SHAPE_MAX_PARALLEL_REQUESTS Maximum number of parallel export shape requests (multi-user mode only). 0 (no limit)
PENPOT_MCP_AUTH_URI Public base URL of the MCP authorization service, e.g. https://penpot.example/api/mcp/oauth. Set to enable OAuth; requires --multi-user. Use HTTPS (HTTP is allowed on loopback for development), without credentials, query parameters, or a fragment. (unset; OAuth disabled)
PENPOT_MCP_REDIS_URI Redis connection URI (e.g. redis://host:6379) enabling multi-instance horizontal scaling via Redis pub/sub task routing (multi-user mode only). When unset, the server runs in single-instance mode, requiring the plugin and MCP client to connect to the same instance. (unset)

Logging Configuration

Environment Variable Description Default
PENPOT_MCP_LOG_LEVEL Log level: trace, debug, info, warn, error info
PENPOT_MCP_LOG_DIR Directory for log files; file logging is enabled iff this is set to a non-empty value (unset)

Plugin Server Configuration

Environment Variable Description Default
PENPOT_MCP_PLUGIN_SERVER_HOST Address on which the plugin web server listens (single address or comma-separated list) (local only)

The Server's Modes of Operation

The above instructions describe how to run the MCP server and plugin server locally, for a single user – its simplest mode of operation.

The server supports additional modes of operation, which are intended for non-local usage (and which are relevant to hosted deployments only). Some of the aforementioned configuration options control these modes.

  • remote mode:
    In remote mode, the server is not assumed to be accessed by a local user on the same machine, with corresponding limitations being enforced (tools offering local file system access are restricted/disabled).

  • multi-user mode:
    In multi-user mode, the server can be accessed by multiple users simultaneously. This mode always implies remote mode. User tokens are passed to the server both when connecting an MCP client and when establishing plugin connections, associating each connection with the respective user and allowing tasks to be routed to the correct Penpot instance. This mode is intended for hosted deployments using the integrated version of the Penpot MCP plugin, which automatically sends the user token to the server when establishing WebSocket connections. It is enabled by running the server with the --multi-user flag.

  • multi-instance mode:
    In multi-instance mode, multiple instances of the MCP server can be run in parallel (load balancing). This can result in connections from the same user being owned by different server instances, making it necessary to dispatch tasks to the correct instance. This is handled through Redis, and the mode is thus enabled by setting the environment variable providing the Redis connection URI.

Development

  • The contribution guidelines for Penpot apply
  • Auto-formatting: Use pnpm run fmt
  • Generating API type data: See types-generator/README.md
  • Versioning: Use bash scripts/set-version to set the version for the MCP package (in package.json).
    • Ensure that at least the major, minor and patch components of the version are always up-to-date.
    • The MCP plugin assumes that a mismatch between the MCP version and the Penpot version (as returned by the API) indicates incompatibility, resulting in the display of a warning message in the plugin UI.
  • Packaging and publishing:
    1. Ensure that the API type data is up-to-date (see above).
    2. Ensure release version is set correctly in package.json (call bash scripts/set-version to update it automatically)
    3. Create npm package: bash scripts/pack (creates penpot-mcp-<version>.tgz for publishing)
    4. Publish to npm: npm publish penpot-mcp-<version>.tgz --access public