anymd

The web context layer for AI agents. anymd reads public web content, normalizes it into structured Markdown, remembers what matters in a private searchable library, and makes it available anywhere through MCP, API and CLI.

The fastest way to try it: put anymd.cc/ in front of a public URL.

curl https://anymd.cc/stephango.com/saw

Connect your agent over MCP:

claude mcp add --transport http anymd https://anymd.cc/mcp

anymd.cc · Docs · API reference · MCP · Pricing

What it does

• Read. A source-aware reader extracts the content of a public URL and normalizes it into structured Markdown with YAML metadata. Dedicated readers handle GitHub, YouTube transcripts, Reddit, Hacker News, X posts, PDFs, Office files, spreadsheets and images.

• Remember. What your agents read while signed in becomes private, reusable context: BM25, full-text, semantic (bge-m3 on Vectorize) and hybrid search, with query fan-out, Reciprocal Rank Fusion and optional Jev tie-breaking. Recall is free.

• Use anywhere. A remote MCP server (read_url, search_library, get_document…) with OAuth 2.1 or API keys, a REST API, a zero-dependency CLI, WebMCP tools in the browser, and the URL prefix: https://anymd.cc/<url> returns Markdown, JSON or an HTML preview.

Related MCP server: Fetchkeep

Responsible by design

• Reads only the URLs a user or their agent requests; no autonomous crawling.

• Public content only; no login, paywall or CAPTCHA bypass.

• Follows robots.txt (anymd token or *), honours domain opt-outs on every channel, and enforces per-site rate limits.

• Libraries are private to each account. Site owners can opt out or request a takedown: Site Owners & Abuse.

Also included

• Scoped access. API keys with presets, capped by role templates.

• Agent-readable site. Every page has a .md twin, plus llms.txt.

• Admin page builder. Typed blocks, revisioned ops, previews and publishing over REST, MCP and CLI.

• Admin control plane over MCP. Owners and admins manage users, credentials, credits, settings, opt-outs, billing health, audit history and system usage through scoped, audited, idempotent tools, with least-privilege presets for admin agents.

• Runs on Cloudflare. One Worker with D1, KV, R2, Vectorize and Workers AI; self-hostable.

Quick usage

# Markdown
curl https://anymd.cc/stephango.com/saw

# JSON
curl -H "Accept: application/json" https://anymd.cc/stephango.com/saw

# With an API key: counted against your credits and saved to your library
curl https://anymd.cc/api/v1/convert \
  -H "Authorization: Bearer $ANYMD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://stephango.com/saw"}'

CLI:

npm i -g https://cdn.anymd.cc/cli/anymd-cli-latest.tgz
anymd https://stephango.com/saw

Claude Code:

claude mcp add --transport http anymd https://anymd.cc/mcp --header "Authorization: Bearer amd_…"

Documentation

The docs source is in src/content/docs/. Maintainer docs: docs/. AI coding agents: AGENTS.md. Reviewers: REVIEW.md.

Self-hosting

anymd deploys as a single Cloudflare Worker:

1. Create a D1 database, two KV namespaces (OAUTH_KV, CACHE), an R2 bucket, and a 1024-dimension Vectorize index with user_id and doc_id metadata indexes.

2. Point wrangler.jsonc at your account, resources and hostnames.

3. Apply migrations/ with wrangler d1 migrations apply.

4. Set the optional secrets you need (names in src/env.ts).

5. npm install && npm run deploy:production.

Full guide: Self-hosting. Upstream CI deploys dev to staging.anymd.cc and main to anymd.cc.

Development

npm install
npm run dev        # build + wrangler dev (staging config)
npm run typecheck
npm test

Contributing

Issues and pull requests are welcome. If a page converts badly, open an issue with the URL. Read AGENTS.md for layout and conventions, and REVIEW.md for what reviews check.

License

MIT. Copyright (c) 2026 Digitop.ai.

Built by Digitop.ai. Contact: hello@digitop.ai.