Docs
Documentation
anymd is the web context layer for AI agents. Docs for MCP, the URL API, REST, CLI, WebMCP, library search and self-hosting.
anymd is the web context layer for AI agents. It reads public web content, normalizes it into structured Markdown, remembers what your agents read in a private searchable library, and makes all of it available through MCP, the API and the CLI.
The fastest way to see it: put anymd.cc/ in front of a link and you get the content back, with no nav bars, cookie banners or scripts. Just text your model can use, with YAML frontmatter on top.
curl https://anymd.cc/stephango.com/saw
The rest of these docs cover doing it from inside an agent, from code and from your terminal, and recalling what was read later.
Pick your path
| I want to… | Start here |
|---|---|
| Try it in 60 seconds | Quickstart |
| Convert a URL with zero setup | URL API |
| Call anymd from my app | REST API |
| Use it from the terminal | CLI |
| Give Claude, Cursor or my agent web context over MCP | MCP server |
| Let in-browser agents use anymd.cc | WebMCP |
| Find something my agents read last month | Library & search |
| Find public posts on X, Facebook, Instagram, Threads, LinkedIn or all of them | Social search |
| Know which sources have dedicated readers | Supported sources |
| Scope a key for CI or a teammate | API keys & roles |
| Understand credits and plans | Billing & credits |
| Run my own copy on Cloudflare | Self-hosting |
Core ideas
- Read, remember, use anywhere. anymd reads a source once, keeps it as private context, and serves it to every agent you connect.
- The URL is the API.
https://anymd.cc/<url>returns Markdown. No key needed for light use (50 reads a day per IP). - One engine, every channel. The URL API, REST, CLI, MCP and WebMCP all run the same conversion pipeline, so output is identical wherever you call it from.
- Pay for new context, not seats. Processing a new source uses credits by complexity (a web page is 1). Searching your library and cached reads are free. See Billing & credits.
- Your library remembers. Sign in (or send an API key) and every source your agents read is saved to a private library with BM25, full-text, semantic and hybrid search.
- Scoped access. Every API key and OAuth grant carries scopes, capped by its owner's role. See API keys & roles.
- Responsible reading. anymd only reads URLs you or your agent request, only public content, and follows robots.txt, domain opt-outs and per-site rate limits. See Supported sources and Site Owners & Abuse.
These docs are Markdown too
Every page on anymd.cc has a Markdown twin. Append .md to the path (/docs/api.md, /pricing.md) or send Accept: text/markdown. Point your agent at them directly.
The machine-readable API spec lives at /api/v1/openapi.json, with an interactive reference at /docs/api/reference.
Get help
- Something converts badly? Open an issue at github.com/digitopvn/anymd. Broken pages are how the engine gets better.
- Everything else: hello@digitop.ai.
anymd is open source under the MIT license and built by Digitop.ai.
Updated 2026-10-10 · Edit on GitHub