monday.com and AI: Is there an MCP server for monday.com?

monday.com is a work platform for projects, tasks and processes, built from boards, items and columns. If you want an AI assistant such as Claude or ChatGPT to work with this data, the first question is: is there an MCP server for it?

In short

Yes. monday.com runs an official, hosted MCP server, available on all plans according to monday.com. It reads and updates data. Read-only works through your own OAuth app.

What monday.com offers itself

The MCP server. According to the developer documentation, the hosted server reads and writes monday.com data and runs at https://mcp.monday.com/mcp. According to the integration guide, it uses Streamable HTTP. SSE is deprecated and not supported. Sign-in uses OAuth or a personal API token. The tool list has more than 60 tools, for example for boards, items, docs, dashboards, forms, automations and agents. According to monday.com, some newer tools run against the dev schema (preview) and can change.

Availability. According to the help article, MCP is available for admins and members on all plans. On Enterprise, admins can allow guests. Viewers have no access. Admins turn AI connectors on or off under “AI governance” and can limit MCP to specific workspaces. According to the product page, MCP currently costs nothing extra. Costs for the AI tool come on top. The product page and two monday.com blog posts (1, 2) say admins must install the MCP app from the marketplace. The help article and the article Get started with monday MCP say MCP comes preinstalled and needs no installation. We could not resolve the conflict.

Clients. According to the list of compatible clients, clients such as Claude, ChatGPT, Cursor, Microsoft Copilot, Gemini and Mistral are pre-approved. According to monday.com, an admin must approve other clients under “Connectors > Dynamic Connectors”. The documentation calls this flow “early access”. For Claude, the help article requires a Claude Pro, Max, Team or Enterprise plan.

Rights. According to the security overview, every call runs on behalf of the signed-in user and follows that user’s permissions. There is no shared service account. According to the security overview, a self-service export of detailed MCP audit logs is not currently available. The server runs write tools, including deletes: the list contains tools that delete automations, agents and form questions. With all_monday_api, an assistant runs any GraphQL query. For read-only, monday.com describes the route through your own OAuth app with narrow scopes, for example only boards:read.

The GraphQL API. The API has one endpoint at https://api.monday.com/v2. According to the rate-limit documentation, MCP calls count toward the daily call limit (resets at midnight UTC): 200 (Free), 1,000 (Standard/Basic), 10,000 (Pro/Trial), 25,000 (Enterprise).

AI in the product. According to monday.com, monday AI includes agents, sidekick and monday vibe, among others. Usage runs on AI credits. There, according to monday.com, agents, sidekick and AI blocks, among others, consume credits. According to the pricing model, customers who joined on or after 6 May 2026 buy credits together with seats (at least 1,000 on Basic, 2,000 on Standard, 3,000 on Pro; Enterprise on request). According to monday.com, AI features are available for paying accounts only, and the model covers the “AI work platform” only, not monday CRM, dev and service. According to the pricing page, AI features start at Basic. According to the AI Feature Catalog, the MCP server consumes no AI credits, only API calls. According to monday.com, it does not use customer data to train AI models and does not allow others to do so. For up to 60 days, monday.com may access data to monitor and improve its services. Admins can opt out. According to monday.com, the model providers work under zero data retention. Some models may retain inputs and outputs for up to 30 days if an admin enables them.

Data location. According to the help article, there is a US, an EU and an APAC region, all on AWS. The region follows the first user of the account and cannot be changed later. The EU region is available for Enterprise, and for Standard and Pro accounts created on or after 23 January 2023. Only on Enterprise is customer data hosted solely in the EU. For Standard and Pro, it may also be processed in the United States. Data such as credentials and usage analytics sits in the US region. Because the headquarters are in Israel, customer data may also be processed there. According to the AI FAQs, monday AI follows the account’s region. The sources do not say whether this also holds for the MCP server.

What the community and monday.com offer as open source

The mondaycom/mcp repository is MIT-licensed and maintained by the monday.com AI team. The package @mondaydotcomorg/monday-api-mcp runs locally with an API token and has a --read-only option according to its README. Its “Dynamic API Tools” are in beta there and incompatible with read-only mode. monday.com recommends the hosted server.

Third parties offer servers as well. According to its README, the CData server runs locally and read-only, but needs a licensed CData JDBC driver.

Integration platforms also offer monday MCP servers, for example Zapier, Pipedream and viaSocket. According to Zapier, an MCP call uses 2 tasks from the plan quota. According to viaSocket, 22 monday actions are available, including “Delete Item”. On Pipedream, the tool list loads only after sign-in. The pages do not show which rights, data region and logging apply. Check their terms separately.

What this means for your company

• The official server changes and deletes data. Decide who uses it, and limit it to specific workspaces where needed.
• If you want read-only, plan for your own OAuth app with narrow scopes. Then turn off the default connector.
• Check your data region under Administration before you rely on EU storage. It depends on the plan.
• MCP calls count toward the daily API limit. On Free, that is 200 calls a day.
• What reaches the AI assistant falls under its terms, no longer under monday.com’s.