monday.com MCP Server
Give AI agents secure access to monday.com's structured data and the tools to act on it, via Model Context Protocol — no custom integrations required.
Evidence shows read-only mode (--read-only), tool filtering by mode, dynamic API tools disabled by default with explicit opt-in, and hosted OAuth plus workspace controls — a real least-privilege foundation, deducted one point because the default local configuration remains full read-write; destructive operations such as delete_item have no user-confirmation mechanism (1); data-flow documentation is thin and the token can be passed via CLI argument, a leak risk (1); a security best-practices section supports sensitive_data_handling at 2; no dependency manifest/audit evidence provided (1); delete_item is documented as permanent with no safeguard, a clear deduction on external_effects (1); no rollback/recovery mechanism is described anywhere (0); attribution is clear (monday.com AI team, MIT, third-party terms clarified) (2).
Error-handling evidence is strong: structured error content, rate-limit circuit breaker, GraphQL context-preserving rethrow, all with thorough unit tests — failure_messages earns 3; CI runs prettier/lint/build/test and GraphQL codegen on PRs, giving self_consistency and dependency_availability 2, but no evidence of test coverage over the core tool logic itself prevented full marks.
Audience and scenarios are clearly described (project-management assistant, reporting, multiple clients) — 3; capability boundaries are partly defined via default/apps modes and the dynamic-tools beta flag but upper bounds per mode are not stated — 2; evidence for trigger precision (guidance on which tool to invoke when) is weak in the provided files — 1; environment requirements are explicit (Node 20+, npm 5.2+) — 2.
Install notes are exceptionally complete (multiple clients, hosted vs local, advanced config, dev guide) — 3; MIT license with a clarification that monday.com Developer Terms apply — 3; repository architecture is reasonable but some documentation lives only as external links — 2; no CHANGELOG and a fixed root version 1.0.0 with publish driven by git-diff detection makes versioning/changelog weak — 1; known limitations are noted only for the dynamic-tools beta — 1; maintainer is named but no explicit maintenance commitment or response path — 2.
Tools return structured content with request IDs aiding debugging — output_usability 2; ready-made tool set plus agent toolkit offers real marginal value over hand-rolled GraphQL integration — 2; hosted-vs-local guidance helps users weigh cost — 2; higher scores withheld absent executed verification.
README claims (tool list, configuration parameters) cross-check against repository structure and CI workflows — 2; many key claims (hosted performance, OAuth) point to external support articles rather than in-repo evidence — 2; marketing language ('fastest, most robust', 'significant advancement') is intermixed with factual statements, weakening fact/inference separation — 1.
- The local default configuration is full read-write; always use --read-only unless writes are explicitly needed.
- delete_item/delete_column are permanent with no confirmation or recovery; constrain them in production via gateway policies or hosted workspace controls.
- Avoid passing the API token via the -t CLI argument; prefer environment variables and rotate tokens regularly.
- Enabling Dynamic API Tools exposes the full GraphQL mutation surface to the agent; assess prompt-injection risk before enabling.
- The publish workflow auto-publishes changed packages to npm based on git diff; verify tag protection and npm 2FA on the supply chain.
- No dependency audit or CHANGELOG is provided; review dependency changes yourself before upgrading.
What does this agent do, and when should you use it?
monday.com MCP is an open-source project maintained by the monday.com AI team, shipping two core packages: `@mondaydotcomorg/monday-api-mcp`, a plug-and-play MCP server, and `@mondaydotcomorg/agent-toolkit`, a toolkit supporting both OpenAI and MCP-based agents. It lets AI clients such as Claude Desktop, Cursor, ChatGPT, Microsoft Copilot Studio, Mistral le Chat, and Gemini CLI read and write boards, items, columns, groups, forms, and user data through the monday.com API. A hosted MCP service at https://mcp.monday.com/mcp offers OAuth authentication, workspace controls, and automatic updates, while the server can also run locally via npx. The server additionally exposes Apps Framework tools and a Beta Dynamic API Tools mode that lets agents generate and execute arbitrary GraphQL queries. It is MIT licensed and requires Node.js v20+ plus a monday.com API token.
The MCP server exposes a toolset over monday.com's GraphQL API: item operations (create_item, delete_item, get_board_items_by_name, create_update, change_item_column_values, move_item_to_group), board operations (create_board, get_board_schema, create_group, create_column, delete_column), account operations (list_users_and_teams), and WorkForms operations (create_form, get_form). Once a client connects — via JSON config such as npx @mondaydotcomorg/monday-api-mcp@latest with a MONDAY_TOKEN env var, or the hosted endpoint https://mcp.monday.com/mcp — you can ask an assistant in natural language to query boards, create items, change column values, or build reports. With --enable-dynamic-api-tools, agents can also call all_monday_api to execute any GraphQL query or mutation dynamically, using get_graphql_schema and get_type_details to explore the API surface. The server supports a --read-only mode and a --mode apps switch for Apps Framework tools.
- A project manager manages monday.com boards from Claude Desktop in natural language: creating tasks, updating statuses, and moving items between groups
- A data analyst has an AI assistant extract data from multiple boards, generate summary reports, or create new reporting boards
- A team admin rolls out AI access company-wide via the hosted MCP's OAuth and workspace controls, limiting access to specific workspaces
- An automation engineer builds custom OpenAI or MCP agents with agent-toolkit for deep integration with internal workflows
- A developer uses Dynamic API Tools to run batch operations or reach advanced monday.com features like docs, workspaces, and activity logs via GraphQL
- A monday.com app developer creates, manages, and deploys custom apps through AI using the Apps Framework tools
What are this agent's strengths and limitations?
- Officially maintained, offering both a hosted service (zero setup, automatic updates, OAuth) and local deployment in the same repository
- Ships a declarative toolset covering items, boards, accounts, and forms, eliminating custom API integration work
- Dynamic API Tools can execute arbitrary GraphQL queries, unlocking the full monday.com API beyond the predefined tools
- agent-toolkit supports both OpenAI and MCP implementations, easing custom agent development
- Core value is tied to the monday.com platform; useless without a monday.com account
- Use of the monday.com API is subject to its Developer Terms, and data access depends on the token's scope
- Dynamic API Tools are in Beta, disabled by default, and incompatible with read-only mode, requiring full API access
- Local runs require a Node.js v20+ environment and manual API token management, adding operational cost versus the hosted mode
How do you install or deploy this agent?
Prerequisites: Node.js v20+ and NPM v5.2.0+, plus a monday.com API token (log in, click your avatar in the bottom-left, select Developers → My access tokens). Running locally requires no package install — configure your MCP client: {"mcpServers": {"monday-api-mcp": {"command": "npx", "args": ["@mondaydotcomorg/monday-api-mcp@latest"], "env": {"MONDAY_TOKEN": "your_monday_api_token"}}}}. Alternatively use the hosted MCP (recommended): add {"mcpServers": {"monday-mcp": {"url": "https://mcp.monday.com/mcp"}}} in Cursor or another MCP client, or install the Monday MCP app from the monday.com marketplace for OAuth. Gemini CLI users can run gemini extensions install https://github.com/mondaycom/mcp or gemini mcp add -t http monday https://mcp.monday.com/mcp.
How do you use this agent?
After configuration, operate in plain language inside Claude, Cursor, or ChatGPT, e.g. "What items do I have in board 123?" or "Can you create a board to manage my project?". Advanced options: --enable-dynamic-api-tools true enables dynamic GraphQL tools (append only to expose just those); --read-only restricts to data retrieval; --mode apps switches to Apps Framework tools; use mcp-remote's --header Api-Version:${API_VERSION} to pin an API version. Security practices: never share your token publicly, use --read-only when you only need data, and rotate tokens regularly.
How does this agent compare with similar options?
The README directly contrasts local deployment with the repository's own hosted service (https://mcp.monday.com/mcp): hosted means no local install, automatic updates, better performance, OAuth, and workspace controls; local or agent-toolkit use suits source-code customization, offline development, or building custom agents.
FAQ
What credentials and runtime are required?
Hosted or local — which should I choose?
Can I restrict the AI to read-only access?
--read-only to enable read-only mode, suitable when you only need to retrieve data.What if the built-in tools aren't enough?
--enable-dynamic-api-tools true); agents can then dynamically generate and execute any monday.com GraphQL query or mutation. Note they are incompatible with read-only mode.