skillbox MCP
skillbox hosts a versioned skill library on your own server and serves it to agents over MCP with scoped access keys
- 280 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server stores client keys and skills that agents later read as instructions.
Install
- Works in
Manual install
git clone https://github.com/kitze/skillbox.gitThen `cd skillbox` and `bash scripts/skillbox.sh setup`. The panel opens at `http://127.0.0.1:4791`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server stores client keys and skills that agents later read as instructions.
This is third-party code. Review the repository files before installing.
What it does
Skillbox is a self-hosted service backed by PostgreSQL. You create skills or import them from GitHub URLs and folders, then build profiles with permission grants and issue client keys. An agent connects over HTTP to `/mcp` or through the stdio bridge `cli/skillbox.mjs`. Base tools are `search_skills`, `recommend_skills`, `load_skill`, `read_skill_file` and `report_skill_use`. A skill is fetched with path, hash and size validation, and no code runs.
Who it is for. Teams whose skills are scattered across laptops and who need one versioned source with separate access for each agent.
Good fit when
- Different agents and people need different skill sets from one store.
- A skill must roll back to an earlier version with a visible history.
- You want to see which skills agents actually load.
- You need protection from a skill that changes after review: fetching verifies hashes.
Not a fit when
- A `.claude/skills` folder in one repository is enough.
- You have no Docker or PostgreSQL and do not want to run a server.
- You need a hosted service with no administration: the project is self-hosted.
Example request
Search skillbox for a code review skill, load it and record the use through `report_skill_use`.MCP
- Transport
- http, stdio
- Authentication
- API key
Environment variables
`SKILLBOX_ADMIN_TOKEN`required, secret- Panel login token of at least 32 random characters.
`SKILLBOX_ORIGIN`required- Server address such as `http://127.0.0.1:4791`.
`SKILLBOX_URL`- Server address for the stdio bridge.
`SKILLBOX_CONFIG`secret- Path to a protected file holding the client key for the stdio bridge.
Limitations
Setup requires Docker Engine or Desktop with Compose v2 and Bash (Linux, macOS, WSL). The panel login uses `SKILLBOX_ADMIN_TOKEN` from the local `.env`, which must stay private. Remote access needs `SKILLBOX_ORIGIN` and usually HTTPS through Caddy. Jev recommendations are optional and need your own OpenRouter, TypeSafe AI or Vercel AI Gateway key. Revoking a key blocks future access but does not retract skills already fetched. `scripts/install-client.py` edits client settings and should be read before running.
How to disable. Remove the `skillbox` entry from the agent MCP config and revoke the client key in the panel. Stop the containers if the server is no longer needed.
FAQ
Does skillbox run code from skills?
No. Fetching validates paths, hashes, size and the executable flag, then writes files atomically and runs nothing.
Do I need my own key for recommendations?
Only for the optional Jev recommendations. Search works without them.
How do I connect a client without HTTP?
Through the stdio bridge: command `node` with arguments `cli/skillbox.mjs mcp` and the `SKILLBOX_URL` and `SKILLBOX_CONFIG` variables.
Which tools does the agent see?
The base ones: `search_skills`, `recommend_skills`, `load_skill`, `read_skill_file`, `report_skill_use`. Write and proposal tools appear according to client permissions.
Can I import skills from GitHub?
Yes, from a public GitHub URL with a commit-pinned preview and a revision conflict check.
Discussion
No questions or reviews yet.
Related
MCP serversA skill registry for teams with security scanning and versioning; the MCP server finds and installs skills on request.
A CLI and MCP server for Agent Skills: spec validation, installs from GitHub and on-demand loading of the right skill
An MCP server that opens SKILL.md skills to clients without native support, such as Codex, Copilot and Cursor
The skillx CLI searches a skills catalog, vets skills before install, pins versions in a lock file and serves an MCP installer.
Skill search from inside the conversation: the agent finds a SKILL.md, reads it and applies it with no prior install
Local semantic search over 374 skills: the agent finds one by describing its task and reads the full text only after choosing.
README in short
A self-hosted skill library with MCP, separate client keys and optional recommendations.