PlugWiki

Convex MCP Server, Convex's own MCP server

OfficialLocal

Lets your AI assistant look inside your Convex backend: tables, data, functions, logs and environment variables.

This is Convex's own server. It ships inside the convex command-line tool, so you start it with the command in the install box and it runs on your machine. By default it works on your development deployment and signs in with your existing Convex login.

Before you use it, know that it can run your functions and change environment variables, not just read data. Production is blocked unless you turn it on.

Good for

  • Convex developers who want Claude, Cursor or Codex to debug with real data and logs
  • Teams tuning slow or conflicting functions, using the health report for the last 72 hours
  • Anyone who wants to query a dev database in plain English without writing a script

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install Convex MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the Convex MCP server to Claude Code. Run this command:
    
    claude mcp add convex -- npx -y convex@latest mcp start
    
    Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js 22 or later, Signed in with npx convex login.

Read-only mode: Add --disable-tools run,envSet,envRemove to the arguments. That leaves only the read tools.

Convex recommends its Claude Code plugin, which starts this server for you. The commands here are the manual route and are adapted from Convex's bare command.

Built from the package data we hold. Not yet tested by us. Source.

Tools

  • Lists your tables with their declared and inferred shapes

    Returns every table in a deployment with the schema you declared and the shape Convex has seen in the stored data. Good for getting oriented in an unfamiliar backend.

    Try: “What tables does my dev deployment have, and what fields are in each?”

    Read · tables

  • Reads a page of documents from one table

    Reads documents from a table in the order you choose, one page at a time. It returns a cursor so the assistant can ask for the next page.

    Try: “Show me the 20 newest rows in the messages table.”

    Read · data

  • Runs a throwaway read-only query you describe in words

    Runs a short piece of JavaScript as a query against your data. Convex describes it as read-only, so it can filter, join and count but not change anything.

    Try: “Count how many users signed up each day this week.”

    Read · runOneoffQuery

  • Shows every deployed function and what it accepts

    Returns each deployed function with its type (query, mutation or action), whether it is public or internal, and its argument and return types.

    Try: “List all public mutations in my deployment and what arguments they take.”

    Read · functionSpec

  • Pulls recent function logs, including errors

    Fetches one batch of recent function runs with their log lines. Set it to failures only to see error messages and stack traces. It does not keep streaming.

    Try: “Show the last failed function calls and their stack traces.”

    Read · logs

  • Reports conflicts and heavy reads from the last 72 hours

    Summarizes health problems in the last three days: mutations that keep colliding on the same document, and functions that read too much data. Each comes with request IDs for debugging. Not available when you limit the server with a deploy key.

    Try: “Why is my app slow? Check for performance problems on the dev deployment.”

    Read · insights

  • Runs one of your deployed functions with arguments you give

    Calls a query, mutation or action you already deployed, and returns its result and logs. A mutation or action can change or delete data, so this is the tool that can write.

    Try: “Run the seedDemoData mutation on my dev deployment.”

    Write · run

  • Lists environment variable names, optionally with values

    Lists the environment variables set on a deployment. It shows names only, unless the assistant asks for values too.

    Try: “Which environment variables are set on my dev deployment?”

    Read · envList

  • Reads the value of one environment variable

    Returns the current value of a single environment variable on a deployment, which can include API keys.

    Try: “What is OPENAI_API_KEY set to in dev? Just check that it is set.”

    Read · envGet

  • Sets or changes an environment variable

    Creates a new environment variable or overwrites an existing one on the chosen deployment.

    Try: “Set SITE_URL to http://localhost:3000 on my dev deployment.”

    Write · envSet

  • Deletes an environment variable

    Removes an environment variable from the chosen deployment.

    Try: “Remove the unused STRIPE_TEST_KEY variable from dev.”

    Delete · envRemove

  • Used by the assistant behind the scenes

    Finds your Convex projects and the deployments you can use

    Looks up the development and production deployments for a project folder. The assistant uses the result to pick which deployment every other tool should act on.

    Try: “Which Convex deployments does this project have?”

    Read · status

Read from the source code

Permissions

high risk · 6 things to watch

High risk

The assistant can run your deployed functions and change or delete environment variables on development deployments, so a bad prompt can change real data. Disable the run and env write tools, or use a deploy key for one deployment, until you trust the setup.

What it can do

Watch out for

  • It can run functions that change or delete data

    The run tool calls any deployed function, including mutations, on every development deployment by default. Disable it with --disable-tools if you only want to look. docs.convex.dev

  • Production flags allow writes to live data

    --dangerously-enable-production-deployments also allows run, envSet, envRemove and reading env values in production. Leave it off unless you need it, and prefer --cautiously-allow-production-pii for read-only access to production data. docs.convex.dev

  • It can delete and overwrite environment variables

    On development deployments the assistant can set or remove variables. Add envSet and envRemove to --disable-tools if that is not wanted. docs.convex.dev

  • It can read secrets stored in environment variables

    envGet and envList (with values) return API keys on non-production deployments by default. Disable both if your dev deployment holds real keys. docs.convex.dev

  • By default it can reach every project on your account

    It uses your login, so the assistant can pick any project you own. The --project-dir flag does not stop that. Set CONVEX_DEPLOY_KEY to a deploy key to limit it to one deployment. docs.convex.dev

  • Production is protected by default

    Without those flags, production allows only read tools that do not expose personal data (insights, tables, functionSpec). docs.convex.dev

Technical details
Turning on read-only--disable-tools run,envSet,envRemove leaves only read tools. On production deployments the server is read-only by default.
Talks toYour Convex deployments
Sign-in and keysUses the credentials from npx convex login, or CONVEX_DEPLOY_KEY when set.
Scopes
  • Your Convex account login (npx convex login), which reaches every project on the account

  • Optional deploy key for a single deployment

Questions

Is the Convex MCP server official?

Yes. It is built and documented by Convex and ships inside its own convex npm package. It is not the separate @convex-dev/mcp package, which is for turning a Convex app into an MCP server of your own.

How do I add the Convex MCP server to Claude Code or Cursor?

Use the install box above for your app. Convex also offers a Claude Code plugin and a Cursor plugin that start the server for you.

Does it work on production deployments?

Not by default. In production it only allows read tools that avoid personal data. Starting it with --cautiously-allow-production-pii adds data, logs and one-off queries. --dangerously-enable-production-deployments adds everything, including changes.

How do I point the Convex MCP server at production or a preview deployment?

By default it uses your development deployment. Add --prod for the project's production deployment, or --deployment <name> for a specific one (a name like joyful-capybara-123, dev, prod or local). Production is blocked unless you also pass --dangerously-enable-production-deployments. Convex's page also lists --preview-name and --deployment-name, which the 1.49.0 package still accepts.

Can I make the Convex MCP server read-only?

Yes. Start it with --disable-tools run,envSet,envRemove, which leaves the tools that only read. Environment values can still be read through envGet and envList, so disable those too if your keys are in dev.

Does it need an API key?

No. By default it uses the account you signed in with using npx convex login. To limit it to one deployment, set CONVEX_DEPLOY_KEY to a deploy key. The insights tool does not work with a deploy key.

Does it run on my machine or in the cloud?

It runs on your machine as a local process started by your AI app, and it needs Node.js 22 or later. It talks to your Convex deployments over the network.

Can it see my other Convex projects?

Yes, by default. It uses your account login, and the assistant can choose any project you can reach. The --project-dir flag only sets a default. A deploy key is the way to restrict it.

Is the Convex MCP server free?

The server itself is free and open source under the Apache-2.0 license. What your Convex deployment costs is set by your Convex plan, not by this server.

Information

updated Nov 2, 2026

Publisher
Official
Package
convex
Registry
npm
Transport
stdio
Auth
token in env
Version
1.49.0
License
Apache-2.0

Convex MCP docsGitHubnpm

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 4
  1. Convex MCP Server · docs.convex.dev
  2. convex CLI: mcp.ts · github.com
  3. convex CLI: MCP tool registry · github.com
  4. convex on npm · npmjs.com

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.