# mcptoon [Verified] [Health: Active]

**Category:** 💰 Finance & Fintech  
**Repository:** https://github.com/activeing123/mcptoon  
**GitHub Stars:** 369  
**Views:** 20  
**Installs:** 0  
**Upvotes:** 1  
**Directory Page:** https://allmcps.com/mcp/mcptoon

## Description
Every Agent discovers all your MCP tools from one install; major token savings via TOON output.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "mcptoon": {
    "command": "uvx",
    "args": ["mcptoon"]
  }
}
```

## Documentation

## What mcptoon does

mcptoon reduces the context required to work with MCP tools, agent skills, and documentation. Instead of placing every full tool schema into an agent context, it provides a compact catalog and fetches a schema when the tool is needed. Tool results can also be compressed and recovered later through a handle, while identical output within a session can be represented by a reference.

The same approach applies to skills and documentation. Skills use a small pointer backed by lookup, and documentation queries can provide a relevant chapter rather than an entire document. The project reports token measurements through its benchmarking command, including accounting based on `tiktoken`.

## How it works

After installation, `quickstart` detects available MCP servers, configures detected agents, and installs the `/mcptoon` skill. Existing server definitions can be found with `discover`, without reinstalling those servers. Plain-language searches use `search`, and `select` can return a smaller set of candidates for a task.

Tools can be inspected only when required with `inspect`, invoked through `call`, or located automatically with `call --auto`. `manifest` produces a compact catalog and can export it in MCP, OpenAI, or OpenAPI formats. The CLI can run directly from shell-capable agents; `sync` writes client configuration for supported GUI applications, and `serve` provides a single MCP server entry.

mcptoon also supports OpenAPI and GraphQL endpoints through dedicated list and call commands. OAuth flows include PKCE, client credentials, and Dynamic Client Registration. MCP resources, prompts, and completion are available through the corresponding commands, and `bake` stores reusable connection definitions while referring to secrets through environment variables or files.

## Setup and configuration

Install the Python package with:

```bash
pip install mcptoon
```

Then run:

```bash
mcptoon quickstart
```

This performs discovery and agent setup in one step. Useful follow-up commands include `mcptoon manifest`, `mcptoon mode`, `mcptoon bench`, `mcptoon skills resolve "<task>"`, and `mcptoon docs resolve "<question>"`. The installation is described as having no external dependencies and requiring no initial configuration.

## Tools and capabilities

- Compact tool catalogs with on-demand schema retrieval
- Reversible result compression and session-level deduplication
- MCP server discovery, registry search, task selection, and headless calls
- Skill and documentation lookup with agent synchronization
- OpenAPI, GraphQL, OAuth, MCP resources, prompts, and completion support
- Manifest export for MCP, OpenAI, and OpenAPI consumers

## Limitations and notes

A fresh installation can remain in direct mode, where it is installed but not connected to an agent and therefore does not reduce that agent’s context. Run `mode` to check the current operating mode. GUI configuration is handled through `sync`, while shell-capable agents can invoke the native CLI directly. The project describes support across more than 100 desktop and CLI agents, with specific client configuration examples for Claude Desktop, Cursor, Cline, Windsurf, VS Code, and Claude Code.

_Full upstream README: https://allmcps.com/mcp/mcptoon/readme_