Log inSubmit server
Mcptoon logo

Mcptoon

mcptoon reduces MCP tool and skill context overhead through compact indexes, on-demand retrieval, and local routing.

InstallRepositoryWebsite
Full README Compare alternatives Embed widget

Install Mcptoon

Pick your client, then add it in one click or paste the config.

STDIOPython
~/Library/Application Support/Claude/claude_desktop_config.json
claude_desktop_config.json
{
  "mcpServers": {
    "mcptoon": {
      "command": "uvx",
      "args": [
        "mcptoon"
      ]
    }
  }
}

Settings → Developer → Edit Config, merge this into mcpServers, then fully quit and reopen Claude Desktop.

Install not yet sandbox-tested— we haven't run this command through our automated sandbox yet. Not a red flag; we're working through the catalog.

Overview

  • mcp
  • developer-tools
  • token-optimization
  • agent-skills

mcptoon MCP server manages MCP tools and agent skills locally, exposing compact indexes so agents do not load every full schema into context. It can discover tools, inspect schemas on demand, execute MCP calls, retrieve compressed results, route skills and documentation, and diagnose connection issues. Install it with pip, then use quickstart to detect MCP servers and configure supported agents. Reach for it when large tool or skill collections are consuming agent context.

Use cases

  • Reduce context overhead from large MCP tool catalogs
  • Search a local registry for an MCP tool
  • Inspect a tool schema only before calling it
  • Route agent requests to skills or documentation chapters
  • Diagnose MCP stdio connections and environments

Key features

  • Compact tool and skill indexes
  • On-demand schema and content retrieval
  • Natural-language MCP tool search
  • Uniform MCP tool execution
  • Offline BM25 routing for skills and docs
  • Local TOON result encoding

Security & health

What it can reach and how reliably it runs — from our own checks, not the maintainer’s claims.

ActiveLast checked 2d ago

Recent health check succeeded.

11/11 checks healthy over the last 45d

Supply-chain signal

No high-severity advisories surfaced by our automated scan.

Critical 0High 0Medium 0Low 0

Scanned 15d ago via OSV.dev · mcptoon (PyPI)

Adoption & maintenance

Factual signals from GitHub, npm, and our automated checks — not a rating.

GitHub stars347Stargazers on the source repository.
Last commit2d agoMost recent push to the default branch.
Directory activity1 upvotesConfig copies, upvotes, and views on AllMCPs.

How Mcptoon works

What mcptoon does

mcptoon MCP server reduces the amount of MCP tool and agent-skill information placed in an agent's context window. Instead of exposing every complete schema up front, it maintains local data and presents a compact name index. Full schemas and skill content remain available through explicit retrieval when an agent needs them.

The project is intended for the Developer Tools category and supports local management across multiple agents. The package is described as approximately 363 KB and can be installed with pip.

How it works

The mcptoon MCP server keeps tool descriptions and skills on disk, then supplies short indexes for selection. A catalog of 255 tools is reported as shrinking from 71,929 tokens to 581 tokens in name-only form; the full schema can be fetched later with inspection. Skill routing uses an offline BM25 matcher, so matching does not require a model API.

Tool results can also be encoded as TOON with the --toon option. The README reports a 5.7% average reduction on a mixed result sample, with results varying by shape. Session deduplication prevents identical output from being repeated during a turn. Full content is not discarded: schemas, skill text, and original results can be retrieved when required.

Setup and configuration

Install the package with:

Terminal
pip install mcptoon

Run mcptoon quickstart to detect MCP servers and configure detected agents. The command also installs the bundled /mcptoon skill into detected agents. Useful commands include mcptoon manifest for the compact tool catalog, mcptoon mode for checking the current operating mode, and mcptoon doctor for diagnostics. The --dir option provides isolated configuration storage for sandboxed environments.

A direct mode is available after installation, but it does not reduce context until mcptoon is connected to an agent.

Tools and capabilities

The mcptoon MCP server provides or documents these operations:

  • search finds MCP tools through natural-language queries against a registry described as containing more than 17,000 entries.
  • inspect retrieves a selected tool's schema only when needed.
  • call provides a uniform, headless way to execute MCP tools.
  • retrieve obtains the original content represented by a compressed handle.
  • doctor checks and repairs stdio pipes and environments.
  • skills resolve routes requests to installed skills.
  • docs resolve selects relevant chapters from indexed Markdown documentation.
  • bench measures token usage with tiktoken accounting.

The bundled agent skill lets supported clients invoke these functions conversationally. The documentation router indexes Markdown trees and supplies targeted chapters rather than an entire documentation set.

Limitations and notes

Token savings depend on the tool, skill, and result shape. The README's result benchmark ranges from 1.8% to 48.6%, so the reported reductions are not universal. A compact index also means an agent may need an additional inspection or retrieval step before calling a tool.

The project is local and offline-oriented, but it does not automatically save context in direct mode. Run quickstart or otherwise connect it to an agent before expecting routing and compression to affect agent turns. Supported client examples include Claude Code, Cursor, Windsurf, Zed, Cline, and Codex; the repository also claims broader support across more than 103 desktop and CLI agents.

Read the full README View source on GitHub

Reviews

No reviews yet — be the first to share how it worked for you.

Frequently Asked Questions about Mcptoon

mcptoon is an MCP server that connects local AI agents to MCP tools and agent skills through compact indexes, on-demand schema inspection, tool search, uniform calls, result retrieval, and diagnostics. It also routes skills and documentation locally, reducing the context needed to work with large collections of tools.

Install it with `pip install mcptoon`, then run `mcptoon quickstart`. Quickstart detects MCP servers, configures detected agents, and installs the bundled `/mcptoon` skill into them.

Yes, mcptoon lists Cursor among its supported agents. Quickstart can detect and configure supported agents, while the bundled `/mcptoon` skill lets Cursor request tool discovery, inspection, calls, retrieval, and diagnostics.

mcptoon reduces usage by presenting compact name indexes instead of loading every full tool schema into context. It fetches schemas only when needed, routes selected skill or documentation content, and can encode tool results as TOON.

Run `mcptoon doctor` to check and repair stdio pipes and environments.

Related MCP servers

All in Finance & Fintech All alternatives
Add a badge to your READMEShow it’s listed — live SVG, Markdown or HTML.

AllMCPs Directory Badge

Full Badge Customizer

Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.

Badge Style:
Live Dynamic SVG PreviewMcptoon AllMCPs Directory Badge
Markdown (GitHub README)
[![AllMCPs](https://allmcps.com/api/badge/mcptoon?style=directory)](https://allmcps.com/mcp/mcptoon)
HTML Embed
<a href="https://allmcps.com/mcp/mcptoon"><img src="https://allmcps.com/api/badge/mcptoon?style=directory" alt="Mcptoon on AllMCPs" /></a>