Convex MCP Server

The Convex MCP Server connects AI assistants such as Claude Code or Cursor to your Convex deployment. Convex is a backend platform that combines a reactive database, serverless functions in TypeScript, and real-time sync between backend and interface in one system. The MCP server makes that deployment reachable for an assistant: instead of opening tables in the dashboard or calling functions by hand, you ask the assistant to inspect data, run a function with test arguments, or find out why a call failed. The server is part of the npm package convex, which according to the registry is licensed under Apache-2.0 (currently version 1.48.0). The related backend repository lives in the GitHub organization convex, has around 12,700 stars, is not archived, and was last updated on October 24, 2026; for the backend itself GitHub reports no standard license, so we only state the license of the npm package here.

What can the Convex MCP Server do?

According to the documentation the server provides several groups of tools. With status you select the deployment the assistant works on. tables, data, and runOneoffQuery show table structure and contents and allow one-off queries. With functionSpec, run, and logs the assistant sees the existing functions, runs one, and reads the logs. insights provides hints about the state of the deployment. Through envList, envGet, envSet, and envRemove environment variables can be read and changed. This is mainly practical in everyday development: the assistant understands not only your code but also the actual data and error messages of the running deployment.

Installation and sign-in

The server runs locally and is started with npx -y convex@latest mcp start; you add this command to your client's MCP configuration. The documentation also recommends the Convex plugin for Claude Code or Cursor, which starts the server automatically. Sign-in uses your global credentials from npx convex login; alternatively you set the environment variable CONVEX_DEPLOY_KEY for a specific deployment. So there is no separate cloud endpoint; the server works through your existing Convex account.

Deliberately securing production

Particularly noteworthy is the tiered security model. By default, access to production deployments is limited to read-only tools without personal data, namely insights, tables, and functionSpec. The flag --cautiously-allow-production-pii adds read-only tools that can reveal user data (data, logs, runOneoffQuery). Only --dangerously-enable-production-deployments allows reading environment variables and write actions such as run, envSet, and envRemove. The --prod flag also requires this approval. This split sets the server apart from simple database connections where access usually means all or nothing.

What is it particularly useful for?

The greatest value lies in developing and debugging. You have the assistant check a schema against real test data, try a new function with sample values, or narrow down an error cause from the logs. Compared with a general database server, this server knows Convex concepts such as function specifications and insights, so the assistant makes better-fitting suggestions.

Security: what to keep in mind

An assistant with write access can run functions and change or delete environment variables. Therefore work with a development deployment first and enable the production flags only deliberately and for a limited time. Retrieved data, logs, and environment variables enter the AI client's context and therefore reach the model provider; this applies especially to user data and secrets. The documentation names neither a license for the server nor prices; we checked those details against the registry and the pricing page.

Costs

The MCP server is part of the free npm package. To use it you need a Convex deployment. According to the pricing page there is an entry plan for personal projects and prototypes that can be used free or extended on a usage basis when needed, a paid Professional plan for the move into production, and Business and Enterprise tiers with SSO, service guarantees, and dedicated deployments. Current terms are on Convex's official pricing page.

Who is the Convex MCP Server for?

It is worthwhile for developers who already work with Convex or are trying a reactive backend platform for a new app and want their assistant to work with real data and logs. Without a Convex project it adds nothing. Start with a development deployment, use read-only tools first, and leave the production flags off until you know the flow well.

Node.js with npx, a Convex account and deployment (sign-in via npx convex login or a deploy key), and an MCP client with stdio support.

With the Convex MCP Server an AI assistant queries tables, functions, and logs of your Convex deployment. This guide leads from sign-in to the first task.

Step 1: Prerequisites

You need Node.js with npx, a Convex account with a deployment, and an MCP client with stdio support, such as Claude Code or Cursor.

Step 2: Sign in

Sign in once with npx convex login; the server uses these global credentials. For a specific deployment you can set the environment variable CONVEX_DEPLOY_KEY instead.

npx convex login

Step 3: Register the server

In Claude Code the command is claude mcp add convex -- npx -y convex@latest mcp start. Alternatively use the Convex plugin for Claude Code or Cursor, which starts the server automatically. Then check with /mcp that the server is connected.

claude mcp add convex -- npx -y convex@latest mcp start

Select a deployment

Have the assistant call the status tool first and choose the desired deployment. Begin with a development deployment.

Securing production

By default only insights, tables, and functionSpec are allowed in production. The flag --cautiously-allow-production-pii adds data, logs, and runOneoffQuery; --dangerously-enable-production-deployments additionally allows environment variables and write actions. Enable these only deliberately.

Step 4: First tasks

Start with reads: "Show me the tables and their schema" or "What do the insights say about my deployment?". Only then have a function run with test values.

Global credentials from npx convex login; alternatively the environment variable CONVEX_DEPLOY_KEY for a specific deployment.

Access to deployments of the signed-in account: tables, data, functions, logs, insights, and environment variables. Production is limited by default to read-only tools without personal data; further rights are unlocked by start flags.

Retrieved data, logs, and environment variables enter the AI client context and therefore reach the model provider. Processing by Convex follows the provider's privacy policy.

Running functions and changing or deleting environment variables when production flags are enabled, exposure of user data and secrets to the model provider, an overly broad deploy key, and prompt injection via stored content.