DeepSeek Harness Plugin

activeing123/dsh-mcptoon

Stars ★ 0 Category Tools & Capabilities Added 2026-09-21

Mounts the mcptoon MCP gateway as one stdio server, so the agent discovers the tools of every MCP server you have configured through mcptoon's compressed view instead of full JSON schemas on each turn. Ships a doctor CLI that checks the mcptoon executable resolves and prints the exact command/args override when it does not.

Install

# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)

dsh plugin --profile web add github:activeing123/dsh-mcptoon

Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).

README

English | 中文

A DeepSeek Harness plugin bundle that mounts mcptoon as an MCP server in one command, so the host agent discovers its MCP tools through mcptoon's compressed view instead of paying full JSON schemas on every turn.

Measured on a 255-tool workspace: tool discovery drops 71,929 -> 581 tokens across 255 tools (-99.2%, measured). The encoding is lossless — mcptoon only changes how tool definitions travel, never what they say.

What you get

After installing, the model sees mcp__mcptoon__<tool> — the same server-qualified naming other MCP hosts use — and every upstream MCP server you have already configured is proxied behind that single stdio connection.

Install

dsh plugin --profile web add github:activeing123/dsh-mcptoon

Or, once the package is on npm:

dsh plugin --profile web add dsh-mcptoon

Then restart the profile (or rely on HMR: editing the entry reconnects the bridge without a process restart).

Requirements

  • Python 3.9+ with the mcptoon CLI on PATH: pip install mcptoon (zero dependencies, installs in seconds).
  • Review the version before you upgrade: pip install "mcptoon==<reviewed>" is the safer default than an unconditional --upgrade.

If the bridge does not come up

The bundle prints a self-check line at activation. When it says not ready, the CLI is not where command points — common on Windows, where pip installs mcptoon.exe next to an interpreter that is not on PATH, often under a path containing a space (C:\Program Files\...).

Get the absolute path — the bundled doctor prints the exact override block:

npx dsh-mcptoon-doctor          # or: node bin/doctor.mjs inside a checkout
python -c "import shutil;print(shutil.which('mcptoon'))"

Then override in your profile's own cordis.patch.ymlone executable in command, arguments in the args array, never a spaced shell string:

- id: mcp-mcptoon
  name: '@deepseek-ai/dsh-mcp-client'
  config:
    serverName: mcptoon
    transport: stdio
    command: 'C:\Program Files\Python312\Scripts\mcptoon.exe'
    args: ['serve']

Sanity-check the composed profile without booting the app:

dsh --profile web --dump-config | grep -A6 mcp-mcptoon

How it works

This package is a bundle: its package.json declares dsh.bundle.patch, and the patch file it points at inserts one entry into the profile's plugin list — an instance of the official @deepseek-ai/dsh-mcp-client bridge configured for mcptoon serve. The bundle only ever inserts; it never rewrites entries it does not own.

Relationship to the Claude Code plugin

mcptoon also ships a Claude Code plugin (plugin.json + SKILL.md + .mcp.json) in the upstream repository. That packaging was the first to bundle MCP management and compression as an installable plugin. This DSH bundle is a second host wrapper, not an endorsement or official support statement from either project.

License

Apache-2.0. mcptoon itself lives in activeing123/mcptoon and is distributed on PyPI.

Content from the project README on GitHub ↗

Links

More in this category

View the whole category →

Community comments

Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.