When a site publishes one, its documentation arrives as a curated index instead of a crawl, which is faster and far less noisy; when it does not, the discovery tool falls back to finding the documentation URLs. Note that the local package and the hosted endpoint expose slightly different tool sets, so pick one and configure against its list rather than assuming both.
A documentation-extraction service with an MCP surface. It checks whether a domain publishes an `llms.txt` file — the emerging convention for machine-readable documentation indexes — fetches it or its full variant, extracts documentation from an arbitrary URL with that support taken into account, and discovers documentation URLs for domains that have not adopted the convention. It also keeps a searchable directory of sites known to publish one.
- Documentation extracted from a URL with `llms.txt` support applied when the site has one — `extract_docs`, `extract_documentation`
- A domain checked for `llms.txt` before you spend anything on crawling it, with the file's details reported — `validate_url`, `verify_llms_txt`
- The file itself fetched, in the short or the full form — `fetch_llms_txt`, `fetch_llms_full_txt`
- Documentation URLs discovered for a domain that does not publish one — `discover_documentation_urls`
- A directory of known `llms.txt`-enabled sites, listed or searched by category or keyword — `list_sites`, `search_sites`
- The same capabilities over REST, including `POST /api/extract` and `POST /api/generate-install`, which generates an install guide from a GitHub repository or a documentation URL
Nothing to authorise. Two ways to connect. The local npm server exposes the six tools above, and the README names it `@llm-energy/mcp-server`, which npm does not have; the manifest calls the published package `llms-forge` 1.0.6. The other is the hosted HTTP endpoint `https://modelcontextprotocol.name/mcp/extract-llms-docs`, whose surface is five tools — `fetch_llms_txt`, `fetch_llms_full_txt`, `extract_documentation`, `verify_llms_txt` and `discover_documentation_urls`. The repository is registered as `io.github.nirholas/extract-llms-docs`.