mcptoon CLI
A zero-dependency Python CLI: one catalog of MCP servers and skills for every agent, with only a compact index in context
- 405 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The sync command writes other agents' configs, and importing servers reads their entries
- Through call and exec the agent calls any connected server and runs code in a sandbox
- It stores connections and OAuth data: secrets are saved by reference to an environment variable or a file
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/mcptoonDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The sync command writes other agents' configs, and importing servers reads their entries
- Through call and exec the agent calls any connected server and runs code in a sandbox
- It stores connections and OAuth data: secrets are saved by reference to an environment variable or a file
This is third-party code. Review the repository files before installing.
What it does
mcptoon keeps MCP servers and skills in one place and hands the agent a compact catalog in place of the full schemas of every tool. A full schema is fetched only when a tool is actually needed. It works as a plain shell command. An agent that can run a shell gets every feature without a separate MCP server or setup. For GUI clients mcptoon sync writes the JSON configs of Claude, Cursor, Cline, Windsurf, VS Code and Claude Code. mcptoon discover finds servers you already have, mcptoon search looks through a registry of 17,000 servers, and mcptoon call calls a tool on the spot. Skills are served the same way: one pointer in place of the text of every SKILL.md, and mcptoon skills sync puts the catalog into every agent's folder. Through OpenAPI and GraphQL mcptoon calls any REST API, with OAuth sign-in built in. In the exec sandbox only stdout enters context, and every call is checked for destructive operations, injection and key leaks. mcptoon serve exposes every server behind one MCP endpoint.
Who it is for. For people who have connected many MCP servers and skills to their agents and want to free context while keeping setup in one place.
Good fit when
- The MCP tool list eats context on every turn and you fear adding more
- Several agents, Claude Code, Cursor and Codex, and editing a JSON config in each is tedious
- You need to find a server by a task description such as "convert HTML to PDF" and call it right away
- You need to call a REST or GraphQL API as a set of commands without writing an MCP server
- You want to measure on your own catalog how many tokens tool descriptions use
Not a fit when
- You have one server with a couple of tools: the saving will be small
- The agent cannot run shell commands and works only through MCP: you would have to connect mcptoon serve as a proxy
- You need a certified security review of servers: mcptoon's checks guard against common mistakes
- You need an always-on cloud service: mcptoon runs locally
Example request
Load the /mcptoon skill and check what MCP tools I currently haveLimitations
After a fresh install mcptoon can sit in direct mode: installed but not wired into the agent, and saving nothing. The mcptoon mode command checks it. The author's figures come from a synthetic sample of 255 tools, and on your catalog the result will differ.
The package is pure Python standard library and installs with pip install mcptoon. It supports 103 clients: 42 run the CLI directly and 61 mount one mcptoon serve entry. What an installed server can do depends on the server itself, and mcptoon neither extends nor limits it.
How to disable. Run pip uninstall mcptoon. The toggle command switches individual servers and tools off without removing them. The /mcptoon skill and the entries mcptoon wrote into agent folders and configs are removed by hand or with sync and skills sync if you used them to place the entries.
FAQ
How do I install mcptoon and connect it to Claude Code?
Run pip install mcptoon, then mcptoon quickstart. The command finds your MCP servers, writes the configuration and installs the /mcptoon skill into every detected agent. After that, ask the agent to load the /mcptoon skill and work through it.
How many tokens does mcptoon save?
On the author's sample of 255 tools the index takes 581 tokens in place of 71,929, which is 99.2 percent less. For skills the gap is 1,572,330 tokens of SKILL.md text against 39 resident and 791 per lookup. The mcptoon bench command prints your own numbers.
How do I find and call an MCP server by task description?
mcptoon install --search github finds and installs a server, mcptoon skills search "make a PDF" finds a skill. The select command picks the few best tools for a task, inspect shows the schema of only the one you need, call calls it, and call --auto finds the server itself.
Does mcptoon replace an MCP server or run alongside?
It runs as a plain shell command, so the agent needs no separate MCP server. If a client cannot run a shell, mcptoon serve exposes every configured server behind one stdio or HTTP endpoint with connection pooling and per-agent keys.
What security checks are built in?
Every call passes checks for destructive operations, prompt injection and credential leaks. Output is compressed reversibly, and mcptoon retrieve returns the original result. mcptoon runs locally and collects no telemetry.
Discussion
No questions or reviews yet.
Related
CLI toolsA CLI that turns any MCP server, OpenAPI spec or GraphQL endpoint into commands at runtime, with no codegen
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
An MCP bridge: a single run_python tool runs code in a container and calls your MCP servers
A Python framework and CLI for MCP: write servers as plain functions, test tools and install servers into Claude Code, Cursor and other clients
A CLI for any MCP server: list and call tools, resources and prompts, run mock and proxy modes, guard access and manage agent configs
A read-only MCP server for the HeyClaude registry: an agent finds skills, MCP servers and hooks and gets ready install commands
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| mcptoon (this entry) | no | 405 | |
| MCP Hub | no | 1,789 | a manager that starts local MCP servers and connects remote ones behind one endpoint, without a compact schema index |
| Aikito | no | 129 | keeps agents' shared context in one place: instructions, skills and MCP, without compressing descriptions |
README in short
The author starts from three problems: tools eat context, setup lives in every agent, and the right server gets lost among thousands. Two paths follow, one for people who use AI tools and one for people who build agents, plus a separate trust and safety section.