Overview

monday.com MCP Server is an open, production-ready Model Context Protocol (MCP) implementation maintained by the monday.com AI team. It provides a standardized HTTP endpoint that lets AI agents and copilot-style tools access monday.com workspace data, call action tools, and receive structured context so the agent can make safe, auditable decisions inside a user’s work OS.

You can use the hosted MCP (recommended) to avoid operational concerns, or run the server locally to customize behavior, extend tools, or develop and test agents offline. The server acts as an adapter between agents and the monday.com API, handling authentication, workspace scoping, and returning data in MCP-compatible formats.

Features

MCP-compliant HTTP endpoint (default path: /mcp) for agent integrations
Secure authentication:
- OAuth flow support for workspace-scoped access
- Personal access token support for dev/testing
Structured context responses (boards, items, columns, users, updates)
Action tools that let agents perform safe mutations (create/update items, post updates)
Agent toolkit package for building agents and tool integrations
Works with multiple LLM ecosystems (OpenAI-compatible agents, Anthropic/Claude-ready)
Hosted service option with automatic updates, reliability and workspace controls

Installation / Configuration

Two common approaches: use monday’s hosted MCP (no local setup) or run the MCP server locally (for customization or development).

Prerequisites:

Node.js v20+
A monday.com account and an API token (or OAuth credentials for production integrations)

Install the MCP package (local usage):

# add the MCP server package to an existing project
npm install @mondaydotcomorg/monday-api-mcp

Minimal local run (typical steps):

# clone the repo if you want the full server source
git clone https://github.com/mondaycom/mcp.git
cd mcp

# install and start in development
npm install
npm run dev

Environment variables (example)

# Personal or app token for monday.com API
MONDAY_API_TOKEN=your_personal_token_here

# If running locally, set port / base url
MCP_PORT=3000
MCP_BASE_URL=http://localhost:3000/mcp

# For OAuth-based deployments
OAUTH_CLIENT_ID=your_client_id
OAUTH_CLIENT_SECRET=your_client_secret
OAUTH_REDIRECT_URI=https://your-host/callback

Common MCP client config (add to your agent / tool settings):

{
  "mcpServers": {
    "monday-mcp": {
      "url": "https://mcp.monday.com/mcp"
    }
  }
}

If you self-host, replace the URL with your server:

"url": "http://localhost:3000/mcp"

Available Resources

GitHub repository (source, issues, contributions): https://github.com/mondaycom/mcp
npm package: @mondaydotcomorg/monday-api-mcp
Hosted MCP (recommended for production) — see monday.com support docs for platform-specific guides (ChatGPT, Claude, Copilot, Cursor)

Included packages you’ll commonly use:

@mondaydotcomorg/monday-api-mcp — the MCP server implementation
@mondaydotcomorg/agent-toolkit — utilities and helpers for building agents that integrate with monday.com

Use Cases

Copilot in an IDE: provide task and board context to an assistant so it can create issues or update task statuses on behalf of a developer.
Chatbot for project management: integrate an LLM-driven agent that can summarize a board, surface blockers, and create follow-up items via action tools.
Automation & scripts: programmatically run agents that take safe, auditable actions (e.g., route a task, escalate by posting an update).
Testing and development: spin up a local MCP server to iterate on custom tools, add specialized transformation logic, or debug OAuth flows before going to production.
Enterprise deployments: use hosted MCP to control workspace access, manage tokens centrally, and rely on monday.com’s uptime and security posture.

Example: Agent flow (high level)

Agent requests context from MCP endpoint.
MCP server authenticates the request (OAuth or token).
MCP gathers structured board/item context and returns it in MCP format.
Agent uses context to decide on an action; invokes a named action tool exposed by the MCP.
MCP validates the action, executes it against the monday.com API, and returns confirmation.

Tips for Developers

Prefer hosted MCP for production deployments to avoid maintaining constant updates and auth flows.
Use workspace-scoped OAuth for multi-user environments; personal tokens are suitable for quick tests and demos.
Build idempotent action tools — agents may retry on network issues.
Log actions and context snapshots for auditing and reproducibility.

Where to go next

Clone the GitHub repo to inspect source and examples: https://github.com/mondaycom/mcp
Read integration guides for specific AI platforms (monday.com support pages) to connect Claude, ChatGPT, Copilot, Cursor, etc.
Explore the agent-toolkit package to streamline building agents that consume MCP context and call action tools.

monday.com MCP Server for AI Agent Integration