Aider MCP server: the community one fails to start

A wrapper that documents fourteen tools and never reaches them, why it breaks, and how to drive Aider from an agent anyway.

Last verified 1 September 2026 · from The Aider repository, the community package on npm and a first-hand run of it

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Aider MCP server, and is it usable?

Aider is a command-line coding assistant that edits files in a git repository. It is an agent in its own right, which makes an MCP wrapper an unusual idea: one agent driving another.

The community wrapper takes that idea seriously. Its README describes starting Aider as a detached background process so that a calling agent can check the chat history later. Its README calls that agentic swarm behavior: one agent starting another and reading its output.

The fourteen documented tools cover that loop: aider_message_prompt starts a run, aider_check_progress reads the chat history and process status, and the rest handle git state, file reads and configuration.

Every one of the 90 vendors in this directory gets the same four checks. Aider passes 2 of them.

Aider works in code, tests and review, and nothing usable connects to it.

Drive Aider as a program, not as a connector. It already takes a prompt on the command line and edits files in place, so an agent with shell access needs no MCP server at all.

What can an AI agent do with Aider?

Nothing, over MCP. Two things explain that, and the first is why this page has no setup steps.

Do you need the MCP server at all?

No. Run Aider directly. It is a command-line program that takes a message and edits files in a git repository. An agent that can run shell commands drives it with no MCP server in between.

That is also what the wrapper does underneath. Its README lists node, aider and git as environment prerequisites, so the executables have to be present either way, and the wrapper adds process management rather than capability.

For the git operations the wrapper bundles, a repository connector covers the same ground with maintenance behind it. GitHub's server exposes branches, commits and pull requests across 85 tools, and Graphite exposes the gt command line directly.

For code review of whatever Aider produces, Semgrep runs locally, and its tool definitions cost about 1,624 tokens of the agent's context, and Codacy and SonarQube reach platform findings.

Give the agent a shell and call Aider directly. Use a repository connector for the git half.

Aider MCP server setup

Nothing here installs an Aider server worth running. The one community package fails at import, so this is the route that actually works.

Aider is a command-line tool. An agent that has a shell can drive it directly, which needs no connector and no package that has to keep working.

The repository half is separate, and there a real server exists.

aider --message "add a retry to the upload path" src/upload.py

claude mcp add-json "github" '{"command":"github-mcp-server",
  "args":["stdio","--read-only"],
  "env":{"GITHUB_PERSONAL_ACCESS_TOKEN":"<your-token>",
         "GITHUB_TOOLSETS":"repos,pull_requests"}}'

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

npx -y mcp-aider-executor@1.1.0
# SyntaxError: does not provide an export named
# 'StdioServerTransport'

Read-only access, permissions and security

• A shell is a wide grant

  Driving Aider directly means the agent can run commands, which is broader than any connector on this page and should be a deliberate choice.

• Aider edits files in place

  It is a write tool by design, so run it against a branch and review the diff rather than trusting the edit.

• Model keys live in Aider's own environment

  Nothing on this page asks you to put them into an MCP configuration.

• --read-only separates inspection from action

  Letting the agent read the repository without write tools keeps the review honest even when Aider itself can write.

• The broken package is not a security finding

  It fails at import and registers nothing, so it exposes nothing; it simply does not work.

Troubleshooting

You cannot find an official Aider MCP server

There is not one. mcp appears in Aider's own repository only as a dependency pin, not as a server it publishes.

`mcp-aider-executor` installs but the server never starts

It fails at import. Running version 1.1.0 on 24 August 2026 raised a SyntaxError: the requested module @modelcontextprotocol/server does not provide an export named StdioServerTransport.

The package pins @modelcontextprotocol/server at ^2.0.0-alpha.2, an alpha line whose exports moved. Nothing in your configuration causes this.

The README lists 14 tools but the agent sees none

The failure happens before any tool registers. Fourteen tools are documented and zero are reachable, so an agent connects to nothing.

npm shows several versions published on the same day

Four versions were published on 15 May 2026, running 1.0.0, 1.0.1, 1.0.2 and then 1.1.0. The latest is 1.1.0, and it is the version that fails.

You want an agent to drive Aider anyway

Aider is a command-line tool, so the working route is the command line. Give the agent a shell rather than a broken connector.

Aider MCP server: Reference

What engineers report

There is no adoption to report. The import failure is reproducible, and that reproduction is the most useful thing on this page.

The Aider repository, the npm registry and the official MCP registry were checked on 24 August 2026. The community package was installed and run on the same date; it exited with a SyntaxError naming a missing export from its own SDK dependency.

Should you connect an agent to Aider?

No. A server exists and is not usable. Best for calling Aider from the agent's shell, which is what the wrapper does anyway.

Strongest fit

Nobody. The only server does not start.

Main advantage

None. The fourteen documented tools cannot be called.

Main weakness

A published package that fails at import, with 0 stars and no vendor behind it.

Operational risk

The failure is silent from the client's side, which costs debugging time.

Frequently asked questions