Feishu/Lark OpenAPI MCP

The recall-developer-documents tool supports two primary transport modes:

1. stdio mode (Default/Recommended): Best for integration with AI tools like Cursor or Claude via standard input/output streams.
2. SSE mode: Provides an HTTP interface based on Server-Sent Events, suitable for web applications or scenarios requiring network access. When using SSE, the endpoint is accessible at http://<host>:<port>/sse.

# Use stdio mode (default)
lark-mcp recall-developer-documents -m stdio

# Use SSE mode on localhost
lark-mcp recall-developer-documents -m sse -p 3000

# Use SSE mode and listen on all network interfaces (remote access)
lark-mcp recall-developer-documents -m sse --host 0.0.0.0 -p 3000

The driveV1 domain provides several tools for interacting with document comments:

• Retrieval: Use drive.v1.fileComment.list to get all comments (paginated, default 50 per page) or drive.v1.fileComment.batchQuery for specific IDs. Use drive.v1.fileCommentReply.list to get replies to a specific comment.
• Creation/Update: Use drive.v1.fileComment.create for global comments and drive.v1.fileCommentReply.update to edit replies.
• Management: Use drive.v1.fileComment.patch to resolve or restore comments, and drive.v1.fileCommentReply.delete to remove replies.

The corehrV2 domain provides tools for managing organizational structure adjustments and the associated approval processes.

Key capabilities include:

• Tracking Adjustments: Query organizational structure adjustment records via corehr.v2.approvalGroups.get. This allows you to check the status of an approval and retrieve specific details regarding department or personnel changes using corehr.v2.approvalGroups.openQueryDepartmentChangeListByIds and corehr.v2.approvalGroups.openQueryJobChangeListByIds respectively.
• Approval Management: Retrieve a specific user's approval task list with corehr.v2.approver.list or manage specific approval tasks (approve/reject) using corehr.v2.processApprover.update.

When running lark-mcp via Docker, keep the following in mind:

• No password required: The container automatically initializes the secrets service, allowing keytar to store tokens securely without interaction.
• Token persistence: Always mount a volume (e.g., -v lark_mcp_data:/home/node/.local/share) to ensure your authentication tokens are not lost when the container stops.

Use these tools to organize and manipulate files within a user's cloud space:

• File Operations:
  • drive.v1.file.createFolder: Create a new empty folder.
  • drive.v1.file.copy: Copy a file to a different folder (asynchronous).
  • drive.v1.file.move: Move a file or folder to a new location.
  • drive.v1.file.delete: Delete a file or folder (moves it to the recycle bin).
  • drive.v1.file.createShortcut: Create a shortcut to a document in another folder.
  • drive.v1.file.list: List all items (files, folders, docs) within a specific folder.
• Metadata & Stats:
  • drive.v1.fileStatistics.get: Get view counts (PV), unique visitors (UV), and likes.
  • drive.v1.meta.batchQuery: Get metadata (title, owner, creation time) for multiple documents using tokens.

For team sharing or server-side deployment, you can run lark-mcp in streamable mode, which provides an HTTP interface.

Step 1: Start the HTTP Server

Run the following command on your server:

npx -y @larksuiteoapi/lark-mcp mcp \
  -a cli_xxxx \
  -s your_secret \
  -m streamable \
  --host 0.0.0.0 \
  -p 3000

Step 2: Configure MCP Client URL

In your MCP clients, connect via the URL instead of a command:

{
  "mcpServers": {
    "lark-mcp": {
       "url": "http://localhost:3000/mcp"
    }
  }
}

Note on OAuth in Streamable Mode: To use user identity in streamable mode, add --oauth and --token-mode user_access_token to the startup command. Currently, OAuth-enabled streamable services only support localhost.

To use streamable mode, you must first start the Docker container with port mapping and the streamable mode flag. Then, configure your MCP client to point to the local URL.

Step 1: Start the container

docker run --rm -it \
  -p 3000:3000 \
  -v lark_mcp_data:/home/node/.local/share \
  lark-mcp:latest mcp -a <your_app_id> -s <your_app_secret> -m streamable --host 0.0.0.0 -p 3000

Step 2: Client Configuration

{
  "mcpServers": {
    "lark-mcp": {
      "url": "http://localhost:3000/mcp"
    }
  }
}

To access private user data (e.g., personal documents, sending messages as a user, accessing calendars), you must authenticate via OAuth.

Step 1: Perform Login in Terminal

Run the following command to start a local server and open your browser for authorization:

npx -y @larksuiteoapi/lark-mcp login -a cli_xxxx -s your_secret

Note: Ensure http://localhost:3000/callback is configured as the redirect URL in your Feishu/Lark app settings.

Step 2: Configure MCP Client

Update your MCP client configuration to enable OAuth and set the token mode to user_access_token:

{
  "mcpServers": {
    "lark-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@larksuiteoapi/lark-mcp",
        "mcp",
        "-a", "cli_xxxx",
        "-s", "your_secret",
        "--oauth",
        "--token-mode",
        "user_access_token"
      ]
    }
  }
}

To use the Feishu/Lark OpenAPI MCP in AI IDEs like Trae or Cursor, you can use the provided installation deep-links or manually configure the MCP Client using a JSON configuration. You must provide your app_id and app_secret in the args field.

Standard Configuration (Tenant Access Token): Use this for general automation where the application acts on its own behalf.

{
  "mcpServers": {
    "lark-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@larksuiteoapi/lark-mcp",
        "mcp",
        "-a",
        "<your_app_id>",
        "-s",
        "<your_app_secret>"
      ]
    }
  }
}

The recall-developer-documents tool supports different communication modes depending on your integration needs:

1. stdio mode (default/recommended): Best for AI tools like Cursor or Claude that communicate via standard input/output.

lark-mcp recall-developer-documents -m stdio

2. SSE mode: Provides an HTTP interface via Server-Sent Events, useful for web applications. The endpoint is available at http://<host>:<port>/sse.

Run SSE on localhost:

lark-mcp recall-developer-documents -m sse -p 3000

Run SSE on all network interfaces (remote access):

lark-mcp recall-developer-documents -m sse --host 0.0.0.0 -p 3000