pi-crw

web_search + web_scrape for the pi coding agent, backed by the crw web scraper.

pi has no native web capability — out of the box its agent can only reach the web by shelling out to curl. pi-crw registers two first-class tools and advertises them in the agent's system prompt, so the model reaches for them instead of curl.

Install

npx github:us/pi-crw install      # or: bunx github:us/pi-crw install

That copies the extension into pi's extensions dir (~/.pi/agent/extensions/crw.ts). Open a new pi session and the agent has web_search and web_scrape. No config needed.

npx github:us/pi-crw status       # is it installed? which backend?
npx github:us/pi-crw uninstall    # remove it

Installed straight from the GitHub repo (no npm registry). For a pinned version use npx github:us/pi-crw#v0.1.0 install, or clone and run node bin/pi-crw.mjs install.

Backends (auto-detected, zero-config)

Environment	Mode	Notes
CRW_API_URL set	HTTP	POST {url}/v1/scrape / /v1/search. fastcrw.com cloud or a local crw serve. Add CRW_API_KEY for cloud.
unset	CLI	Spawns the crw binary (CRW_BIN, else crw on PATH).

If no backend resolves, the tools are simply not registered — pi keeps working exactly as before. Other tunable: CRW_TIMEOUT_MS (default 60000).

HTTP web_search needs the target server to have SearXNG configured. fastcrw.com cloud bundles it; a self-hosted crw serve needs the SearXNG sidecar. The CLI backend works out of the box.

License & the crw boundary

This package is MIT. crw itself is AGPL-3.0. pi-crw bundles no crw source code — it only invokes a separately distributed crw binary (subprocess) or a crw HTTP endpoint. The boundary is a stable CLI/HTTP protocol, never linking or bundling, so the MIT package and AGPL crw stay cleanly separate.

Development

bun install
bun run typecheck
bun run test            # deterministic: mock crw binary + mock HTTP server

See .claude/CLAUDE.md for the project's conventions and the license rationale.