bathys-mcp (npm wrapper)

Unified local deep-research search service for AI agents. This npm package is a thin installer for the real thing: the Python bathys package (PyPI, GitHub) — one pipeline (search → extraction → query distillation → cache), zero cloud quotas, MCP stdio server with native harness integration.

Install

npm install -g bathys-mcp

The postinstall step installs the Python package (pip install bathys) when Python ≥3.10 is available, then you wire your harness:

bathys-mcp install     # auto-detect harnesses (zcode, Claude, Cursor, VS Code
                       # family, Gemini CLI, Windsurf, Zed, opencode, goose,
                       # Hermes) and register the MCP server — idempotent,
                       # timestamped backups
bathys-mcp doctor      # stack diagnostics
bathys-mcp             # run the MCP server over stdio
bathys-mcp print-config  # manual-wiring snippets for every harness

No Python? Install Python ≥3.10, then pip install bathys and bathys install.

What you get

• 4 MCP tools: deep_research, web_search, read_url, read_urls
• 3 strategy prompts: bathys_deep_research, bathys_source_audit, bathys_fresh_scan
• built-in instructions playbook + tool annotations (harness knows how to use Bathys out of the box)
• researcher subagent profile + skills (--with-agent)
• robots.txt ethics, token budgets, SQLite cache, metrics journal

Full documentation: github.com/Korrnals/bathys/tree/main/docs

License: MIT