# DecimalAI Skills Registry [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/decimal-labs/decimalai-mcp  
**GitHub Stars:** 1  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/decimalai-skills-registry

## Description
Search agent skills ranked by measured lift vs a no-skill baseline, with safety and rankings.

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

```json
"mcpServers": {
  "decimalai-skills-registry": {
    "command": "uvx",
    "args": ["decimalai-mcp"],
    "env": {
      "DECIMAL_API_KEY": "",
      "DECIMAL_API_URL": ""
    }
  }
}
```

**Requires environment variables:** `DECIMAL_API_KEY`, `DECIMAL_API_URL` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What DecimalAI Skills Registry MCP server does

The DecimalAI Skills Registry MCP server gives MCP clients read-only access to the public DecimalAI skills registry. It is intended for agents or developers comparing skills based on measured results, safety information, and ratings rather than popularity or download volume.

The server provides three tools. `search_skills` performs hybrid keyword and semantic searches, with optional category, sort, and result-limit parameters. `get_skill` returns a complete record for a skill slug, including trust and safety scan status, verified benchmark lift, SkillScore, ratings, and the contents of its `SKILL.md`. `get_leaderboard` returns ranked results using views such as `skill_score`, `biggest_improvement`, `efficiency`, or `top_rated`.

## How it works

The DecimalAI Skills Registry MCP server runs as an MCP process and reads from public DecimalAI HTTP endpoints. Searches use the registry skills endpoint, individual records come from the skill-detail endpoint, and leaderboard requests use the leaderboard endpoint. Category-filtered leaderboard requests can use the browse endpoint's documented ranks view because the dedicated leaderboard endpoint is not categorized.

All three exposed tools are read-only. They do not install, edit, or publish skills. Without credentials, the server returns public registry information. Supplying `DECIMAL_API_KEY` allows the same tools to include organization-specific installation information such as `installed_as`; the key is not required for anonymous access.

## Setup and configuration

Install the Python package with `pip install decimalai-mcp`, or run it without a separate installation through `uvx decimalai-mcp`. Python 3.10 or newer is required. The development module can also be started directly with `python -m decimalai_mcp.server`, and the server communicates over stdio.

For Claude Code, register the process with `claude mcp add decimalai -- uvx decimalai-mcp`. Claude Desktop can use a `decimalai` entry in `claude_desktop_config.json` with `uvx` as the command and `decimalai-mcp` as its argument. Leave out the environment section for public anonymous access, or provide `DECIMAL_API_KEY` for organization enrichment.

`DECIMAL_API_URL` defaults to `https://api.decimal.ai` and can point to a self-hosted or local backend. `DECIMAL_API_KEY` is optional and is obtained from the DecimalAI settings page.

## Tools and capabilities

The DecimalAI Skills Registry MCP server exposes these capabilities:

- Search the public registry with keyword and semantic matching.
- Filter or sort search results by supported query parameters.
- Retrieve full skill metadata by slug.
- Inspect trust and safety-scan status.
- Compare verified benchmark lift, SkillScore, and ratings.
- Browse leaderboards for skill score, improvement, efficiency, or top ratings.
- See organization installation identifiers when an API key is configured.

## Limitations and notes

The server deliberately exposes a public, read-only registry surface. It does not provide the authenticated manifest-impact or regression-check operation, because that endpoint evaluates an organization's production traces against a candidate manifest. The documented alternative for CI use is the separate regression-check GitHub Action.

The registry endpoints used by the server are public, but organization-specific enrichment depends on a valid `DECIMAL_API_KEY`. The leaderboard endpoint itself is uncategorized, so category filtering uses the browse endpoint's ranks mode instead. The project is licensed under MIT.

_Full upstream README: https://allmcps.com/mcp/decimalai-skills-registry/readme_