Skip to content
MCP server

ask-kit

MCP server giving coding agents one ask tool that a cheap model answers from code and past

Runs locally
Install

ask-kit is an open-source MCP server for agent tooling. It runs locally on your machine and is started by your AI client.

Pricing
Open source
Platforms
CLI, API, Linux
Runs
Locally
Version
0.3.0
Listing updated

Similar MCP servers

Install

Run in a terminal

claude mcp add ask-kit -- npx -y @temet-ai/ask-kit

Add to claude_desktop_config.json

{
  "mcpServers": {
    "ask-kit": {
      "command": "npx",
      "args": [
        "-y",
        "@temet-ai/ask-kit"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "ask-kit": {
      "command": "npx",
      "args": [
        "-y",
        "@temet-ai/ask-kit"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "ask-kit": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@temet-ai/ask-kit"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "ask-kit": {
      "command": "npx",
      "args": [
        "-y",
        "@temet-ai/ask-kit"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "ask-kit": {
      "command": "npx",
      "args": [
        "-y",
        "@temet-ai/ask-kit"
      ]
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

About ask-kit

The README says the agent gets a single ask tool and the detailed searching happens elsewhere, so the calling agent's own context keeps room for the change it is making. Each question is handled by an investigator, a cheap model with a JavaScript REPL that writes small programs over the data, runs them, and prints only the result back.

The product ships two built-in domains: code, which answers about the repository you are in with path:line citations, and sessions, which searches transcripts already written by Claude Code, Codex, GitHub Copilot CLI, Gemini CLI and opencode. Users can add their own domains as folders with instructions and scripts.

It suits developers who run coding agents heavily and want to cut the context spent on search and file reading. The model can be an OpenAI-compatible API, a subscription-backed CLI, or a local model through Ollama or LM Studio. The project is MIT licensed, and the page says it has no release yet.

Tools this server provides (1)

Taken from the project documentation; the running server may differ.

ask
Answers a plain-language question about code or past sessions, with an optional domain or backend, and continue for follow-ups.

What it does

Single ask tool
Exposes one MCP tool that takes a plain-language question and returns a short answer with its evidence.
Disposable investigator REPL
Each step runs as a fresh JavaScript program in a sandbox, and only printed output returns to the model.
Code and session domains
Built-in domains cover the current repository with path:line citations and past agent transcripts across five tools.
Custom domains
Users can add domain folders with their own data access, instructions and scripts.
Flexible model backends
Works with an OpenAI-compatible API, a existing Claude Code, Codex or Copilot CLI subscription, or a local Ollama or LM Studio model.
Follow-up questions
With continue set to true, a follow-up reaches the investigator that still holds the earlier results.

Good for

  • Finding where a pattern or module is defined in a large codebase without filling the main agent's context
  • Recalling what was decided or tried in earlier coding sessions
  • Checking how existing code is structured before adding a similar feature
  • Searching transcripts across several coding agents for a specific piece of work

Pricing

Open source

Free and MIT licensed; model usage is paid separately to the chosen provider.

Prices change. Check temet.ai before you buy.

ask-kit FAQ

Which agents can use ask-kit?

The README says any agent that accepts MCP servers can call it, and it names Claude Code, Codex, GitHub Copilot CLI, Gemini CLI and opencode.

Does ask-kit need a server or database?

No. The README says it reads transcripts where the agents already save them, with no proxy or database.

What does ask-kit send to the model?

The README says it sends the question and the excerpts the investigator reads to the model the user chose, and that there is no telemetry.

What is the licence?

The website and README both state MIT.

Is ask-kit free?

Yes. ask-kit is open source, so you can use and self-host it for free.

How do I add ask-kit to Claude Code?

Run this command in your terminal: claude mcp add ask-kit -- npx -y @temet-ai/ask-kit

What are the best alternatives to ask-kit?

The closest alternatives to ask-kit are The Context Company, Agentrica, Agentic Platform, Context First MCP and Agent Never Give Up.

Where this comes from

  • Description: written from the project's README, the project's website and its MCP Registry entry.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers