Coda MCP Server
Connect AI agents to your Coda docs, tables, and pages over the official remote MCP server.
Updated
What is Coda MCP?
The Coda MCP server is a Model Context Protocol server, hosted by Coda, that lets AI clients read and write your Coda docs, pages, and tables using plain-language prompts. It exposes Coda as a set of tools an AI agent can call directly, so the agent can find a doc, analyze a table, or update content without you copy-pasting between apps.
The server is remote and Coda-hosted at https://coda.io/apis/mcp — there is nothing to install or self-host. Connecting an MCP-compatible client (Claude, Cursor, ChatGPT, and others) gives the agent running in that client access to Coda's tools scoped to your account.
Authentication uses OAuth 2: when you connect over OAuth the scope is automatically set to read and write. If you want to restrict the agent to read-only or write-only, Coda recommends connecting with a Personal Access Token instead, which lets you pick the scope. The Coda MCP is currently in beta, so the available tools and behavior may change.
Connect the Coda MCP server
Claude Code
- 1
Add the remote MCP server
Run: claude mcp add --transport http coda https://coda.io/apis/mcp
- 2
Authenticate with Coda
Run /mcp inside Claude Code and complete the OAuth login when prompted. Over OAuth the scope is set to read + write automatically.
- 3
Verify the connection
Run /mcp again to confirm Coda shows as connected, then ask the agent to read or update one of your Coda docs.
Cursor
- 1
Open MCP settings
In Cursor, go to Settings > MCP (or Tools & Integrations) and choose to add a new MCP server.
- 2
Add the Coda server
Add an HTTP server with URL https://coda.io/apis/mcp (e.g. in mcp.json: an entry with "url": "https://coda.io/apis/mcp").
- 3
Complete OAuth and use it
Approve the OAuth login to Coda, then enable the server. The agent can now read and write your Coda docs from chat.
Sample use cases
An agent creates a new Coda doc with structured tables from a project kickoff template.
Project setup automated — tables, views, and initial data populated without manual doc creation.
A reporting agent queries Coda tables for weekly metrics and formats a summary.
Automated status reports pulled from the team's existing Coda workspace, delivered on schedule.
An agent adds rows to a Coda tracking table when new tasks come in from another system.
Cross-system task intake — items from email, Slack, or forms land in Coda tables automatically.
A QA agent validates that required fields in a Coda table are populated before a deadline.
Pre-deadline data completeness checks catch missing entries before they block downstream processes.
An agent reads a Coda decision log and summarizes recent decisions for stakeholders.
Stakeholders get concise decision summaries without scrolling through the full doc history.
A planning agent updates sprint columns in a Coda table based on standup notes.
Sprint boards stay current with status changes applied from natural-language standup summaries.
An agent archives completed rows from active Coda tables to a historical archive doc.
Active tables stay clean while historical data is preserved in a separate, searchable archive.
Security & permissions
The Coda MCP server authenticates with OAuth 2. When you connect over OAuth, the granted scope is automatically read and write across the Coda docs your account can access — the agent can both read and modify pages, tables, and rows. To limit exposure to read-only or write-only, connect with a Coda Personal Access Token instead, which lets you choose the scope. Access also respects your Coda plan: Editors and free-plan users get only read-type tools under the beta "free taste," while Doc Maker actions require a paid plan. Review the granted scope before connecting and revoke access in Coda's account settings when no longer needed.
Using Coda MCP with Gamut
In a Gamut workflow, the Coda MCP server becomes a tool your agent can call on an event- or schedule-triggered run — Gamut manages the OAuth connection to Coda for you. For example, a scheduled agent can read rows from a tracking table each morning, summarize status, and write an update back into a Coda page; or an event-triggered agent can append a new row when an upstream system fires. Gamut handles invoking the server and passing results between steps; the agent's access is bounded by the scope granted when you connect Coda.
Frequently asked questions
What is the Coda MCP server?
It is Coda's official, Coda-hosted Model Context Protocol server at https://coda.io/apis/mcp. It lets AI clients like Claude, Cursor, and ChatGPT read and write your Coda docs, pages, and tables using plain-language prompts. It is currently in beta.
Is the Coda MCP server free?
It is in beta. Doc Makers on paid Coda plans get full access. Free plans and Editors get a limited 'free taste' — roughly 30 requests per week (max 60 per month) and read-only tools, with Doc Maker (write) actions requiring a paid plan.
How do I connect the Coda MCP server to Claude or Cursor?
Add it as a remote HTTP MCP server using the URL https://coda.io/apis/mcp, then complete the OAuth login. In Claude Code use 'claude mcp add --transport http coda https://coda.io/apis/mcp'; in Cursor add an HTTP server with that URL in your MCP settings and approve the OAuth prompt.
Is the Coda MCP server hosted or self-hosted?
It is hosted by Coda — a remote server at https://coda.io/apis/mcp with nothing to install. (Several unofficial community MCP servers for the Coda API exist on GitHub for self-hosting, but the official one is Coda-hosted.)
Can Coda MCP agents write data or only read?
Doc Makers on paid Coda plans can both read and write — creating docs, adding rows, updating cells, and modifying pages. Free-plan users and Editors get read-only access with a limited request allowance.
What Coda objects can agents access through MCP?
Agents can work with docs, pages, tables, rows, columns, and formulas. The MCP server mirrors the Coda API surface, so anything you can do in the Coda API is accessible as an MCP tool.
Is the Coda MCP server in beta?
Yes, as of mid-2026 the Coda MCP server is in beta. Core read/write functionality works, but some features and rate limits may change as Coda iterates toward general availability.
Can I use Coda MCP alongside other MCP servers in one agent?
Yes. The Coda MCP server follows the standard protocol, so agents can combine it with other MCP servers — for example, reading from Coda and posting results to Slack or Google Sheets in the same workflow.