Aider is the CLI-first AI pair programmer. It works from a git repo and makes commits as it edits, which suits a tight feedback loop. People searching for an Aider MCP setup usually want one of two things: Aider calling external tools, or Aider running on a local model. Only the second works today.

Does Aider support MCP?

No. As of September 2026 (Aider v0.86), Aider has no MCP client: its options reference lists no MCP option and its release history never mentions MCP. MCP support is an open request in the Aider issue tracker. Aider cannot register ToolPiper as an MCP server or call its MCP tools.

Guides that show an mcp-servers block in ~/.aider.conf.yml or an --mcp-server flag describe options that Aider does not read. Projects named "aider-mcp-server" go the other way: they expose Aider as a tool inside another MCP client, such as Claude Code.

How do I connect Aider to ToolPiper?

Use ToolPiper's OpenAI-compatible endpoint. Aider can talk to any OpenAI-compatible API, so set OPENAI_API_BASE to http://127.0.0.1:9998/v1, set OPENAI_API_KEY to a ToolPiper developer token, and start Aider with an openai/ model name.

  1. In ToolPiper, open Security in the sidebar and create a token with the chat and tools:read scopes. It starts with tp_. Developer tokens are free on every tier.
  2. Load a local model in ToolPiper.
  3. List the model ids ToolPiper serves:
curl -H "Authorization: Bearer $TOOLPIPER_TOKEN" http://127.0.0.1:9998/v1/models

Then start Aider against the endpoint, prefixing the model id with openai/ as Aider's OpenAI-compatible guide describes:

export OPENAI_API_BASE=http://127.0.0.1:9998/v1
export OPENAI_API_KEY=$TOOLPIPER_TOKEN
aider --model openai/<model-id>

The token is required. ToolPiper's /v1 routes answer 401 without a bearer, so a placeholder key such as not-needed fails. Keep the token out of files you commit: an environment variable is safer than a line in .aider.conf.yml.

What does this setup give Aider?

A local model with no per-token bill. Aider's prompts and your code go to the model ToolPiper runs on your Mac instead of a cloud API. Aider keeps its own file-edit, shell and git workflow; it does not gain ToolPiper's tools such as screenshots, browser automation or transcription.

Pick a model that is good at following Aider's edit formats. Small local models often struggle with Aider's diff format, so check Aider's model warnings and try the whole-file edit format if edits fail to apply.

How do I use ToolPiper's tools next to Aider?

Run an MCP client beside it. Claude Code, Cursor, Zed and the other MCP clients connect to ToolPiper over Streamable HTTP at http://127.0.0.1:9998/mcp, so you can keep Aider for repo edits and use one of them when a task needs screenshots, browser tests or other ToolPiper tools.

For Claude Code the setup is one command:

claude mcp add -s user --transport http toolpiper http://127.0.0.1:9998/mcp

See MCP Setup on Mac for the per-editor steps.

What if Aider can't reach ToolPiper?

Three checks: ToolPiper is running, the token is accepted, and the model id matches. Each rules out a different failure.

  1. ToolPiper running. curl http://127.0.0.1:9998/health returns JSON. Connection refused means ToolPiper isn't open.
  2. Token accepted. The /v1/models call above returns a model list. A 401 means the token is missing or mistyped; a 403 means it lacks the chat or tools:read scope.
  3. Model id. The name after openai/ must match an id from /v1/models exactly.

For MCP setup in the editors that support it, see MCP Setup on Mac.