skillinfer-mcp

MCP server for skillinfer — Bayesian skill inference for AI agents and humans.

Lets any MCP-compatible AI agent (Claude, Cursor, VS Code Copilot, etc.) build and query skill profiles via tool calls.

Install

pip install skillinfer-mcp

Or run directly without installing:

uvx skillinfer-mcp

Configure

Add to your Claude Desktop / Claude Code config:

{
  "mcpServers": {
    "skillinfer": {
      "command": "uvx",
      "args": ["skillinfer-mcp"]
    }
  }
}

Tools

Populations

Profiles

Task matching

Persistence

Example session

Agent: load_dataset(name="onet", dataset="onet")
→ Population 'onet' loaded: 894 entities x 120 features.

Agent: create_profile(name="alice", population="onet")
→ Profile 'alice' created (120 features, prior=population mean).

Agent: observe(profile="alice", skill="Skill:Programming", score=0.92)
→ Observed Skill:Programming=0.920 on 'alice'. Observations: 1.
  Top impacted features:
    Skill:Programming: +0.3891
    Knowledge:Computers and Electronics: +0.1842
    ...

Agent: most_uncertain(profile="alice", k=3)
→ [{"feature": "Skill:X", "mean": 0.51, "std": 0.12}, ...]

Agent: predict(profile="alice", skill="Knowledge:Mathematics")
→ {"mean": 0.68, "std": 0.09, "ci_lower": 0.50, "ci_upper": 0.86, ...}