Convex MCP Server

Content

The Convex MCP Server is the official Model Context Protocol server from Convex, built directly into the Convex CLI. It gives LLMs and agentic clients introspection and query access to a Convex deployment — listing tables, paginating data, running deployed functions and read-only one-off queries, reading logs and function specs, surfacing query insights, and managing environment variables.

The server ships in the convex npm package (v1.42.1, Apache-2.0) and runs with npx -y convex@latest mcp start over stdio. It connects to your Convex deployment using your existing CLI login, targets the development deployment by default, and blocks production deployments unless you explicitly opt in — a deliberate safety measure. Full documentation lives in the Convex MCP docs.

Source Review

The following real sources were fetched and reviewed for this entry:

• Convex MCP docs — the mcp start command, deployment-selection flags (--prod, --preview-name, --deployment-name, --env-file, --project-dir), the production-deployment safety gate, --disable-tools, and the full tool list.
• npm: convex — package name and version (1.42.1) and Apache-2.0 license; the mcp subcommand ships in this package.
• get-convex/convex-backend — the official Convex open-source repository.

Repository facts confirmed at review time: official get-convex organization, not archived, actively maintained, with the MCP server distributed via the convex CLI package v1.42.1 on npm.

Features

• Deployment selection — status finds available deployments and returns a selector used by the other tools; flags choose dev, preview, named, or (opt-in) production deployments.
• Table & data introspection — tables lists tables and metadata; data paginates documents in a table.
• Read-only queries — runOneoffQuery runs sandboxed, read-only JavaScript queries against the deployment's data.
• Function tools — run executes deployed Convex functions and functionSpec returns function metadata.
• Observability — logs reads deployment logs and insights surfaces query-optimization information.
• Environment variables — envGet, envList, envSet, and envRemove read and manage deployment env vars.
• Production safety gate — production deployments are inaccessible unless --dangerously-enable-production-deployments is passed.
• Tool disabling — --disable-tools (e.g. data,run,envSet) restricts which tools the server exposes.
• Zero-clone install — runs via npx -y convex@latest mcp start with no separate install step.

Installation

Run the MCP server from the Convex CLI (from within a Convex project):

npx -y convex@latest mcp start

Add it to an MCP client (e.g. Cursor / Claude Desktop):

{
  "mcpServers": {
    "convex": {
      "command": "npx",
      "args": ["-y", "convex@latest", "mcp", "start"]
    }
  }
}

Useful flags:

• --project-dir /path/to/project — resolve deployments from a specific project directory.
• --deployment-name <name> / --preview-name <name> — target a specific or preview deployment.
• --prod with --dangerously-enable-production-deployments — target production (use with care).
• --disable-tools data,run,envSet — restrict which tools are exposed.

Use Cases

• Schema and data exploration — let an agent enumerate tables and page through documents to understand a deployment.
• Read-only analytics — run sandboxed one-off queries to answer questions without mutating data.
• Function debugging — inspect function specs, run functions, and read logs during development.
• Query optimization — use insights to surface slow or problematic query patterns.
• Environment management — review and update deployment environment variables from an agentic workflow.

Safety and Privacy

• Production is opt-in. The server refuses production deployments unless --dangerously-enable-production-deployments is set; enabling it allows reading and modifying live data.
• Some tools mutate. run can invoke mutations and envSet/envRemove change configuration; runOneoffQuery is read-only by design.
• Scope with flags. Use --disable-tools and keep the server on a development deployment when an agent acts autonomously; note --project-dir does not fully prevent access to other projects.
• Data flows to the model. Tables, documents, logs, and function specs are returned to the LLM/MCP client and can include PII; env-var tools can reveal secrets.
• Credential hygiene. The server uses your local Convex CLI login and project directory — keep credentials and any custom --env-file out of version control.

Duplicate Check

No existing entry in the directory matches this server. A search of all directory entries for "convex" across slugs, titles, repository URLs, and install commands returned no results, and there is no prior entry pointing at github.com/get-convex/convex-backend or the Convex CLI mcp subcommand. This is therefore a net-new, non-duplicate entry.

• Source repositorygithub.com 2026-07-08T20:36:14+00:00
• Documentationdocs.convex.dev
• Websiteconvex.dev
• Submitted by davion-knight2026-07-08