dsh-firecrawl

The built-in web_search and web_fetch tools in DeepSeek Harness rely on a single default engine and scraper, so time-sensitive queries often return stale results and heavily rendered pages arrive as messy HTML. dsh-firecrawl reroutes both capabilities through Firecrawl: searches now surface the freshest sources (including news items), and page fetches deliver clean, ready-to-read Markdown capped at 200 000 characters to keep your context window under control.

Under the hood, one server-side module injects into the ctx.web seam and registers two providers. The search provider (id: firecrawl) calls POST /v2/search and merges web and news data into a unified source list with url, title, snippet, and publishedAt. The fetch provider (id: firecrawl-fetch) calls POST /v2/scrape with formats set to markdown and trims the body on overflow. Configuration is minimal: set the API key via an environment variable or a literal string, point baseURL at a self-hosted instance if needed, tune searchLimit between 1 and 20, and adjust the per-call timeout. No changes to the Harness core are required.

This plugin is a good fit for researchers and small teams working inside DeepSeek Harness who need reliable, up-to-date web search and clean Markdown extraction for documentation review or competitive analysis. Firecrawl serves low-quota anonymous traffic out of the box, so a quick trial needs no signup; for production use, grab a free key from firecrawl.dev and set it in an environment variable before starting the profile. One caveat: once activated, each capability routes through a single provider, and PDF pages are not yet supported.