# Turn the web into context your agents can use.
anymd reads public web content, extracts what matters, turns it into structured Markdown, and keeps it in a private searchable library your agents can access over MCP, API or CLI.
- Open source
- Self-hostable
- Respects robots.txt
- No paywall or login bypass
Raw HTML is noisy agent context.
A typical article page is mostly packaging: menus, trackers, pop-ups, related posts, inline styles. Feed that to a model and your agent burns tokens, context and accuracy on cookie banners.
Fewer tokens
The converter shows exactly how much smaller the Markdown is than the page it came from — often 10× or more.
Better answers
Headings, lists, tables, code and footnotes survive. Models reason over structure, not soup.
Seconds, not scripts
No headless browser to babysit. One request, cached at the edge, the same result every time.
Read. Remember. Use anywhere.
One context layer between your agents and the public web: it reads a source once, normalizes it, and keeps it ready for every agent you run.
- 01
Read
A source-aware reader extracts the content and normalizes it into structured Markdown with metadata: headings, links, tables, code and footnotes intact, clutter gone.
read_url → extract → normalize - 02
Remember
Every source your agents read while signed in becomes private, reusable context, indexed for keyword, phrase and meaning search.
library · bm25 · semantic - 03
Use anywhere
Agents reach the same context over MCP, apps over REST, pipelines over the CLI. Export Markdown whenever a human wants it.
MCP · API · CLI
Your agents shouldn’t read the same web twice.
Every source they read can become reusable private context — searchable by keyword, phrase, meaning, or through MCP. Recall is free: searching the library never costs credits.
- BM25Classic relevance ranking over titles, text, domains and tags.
- Full-textExact "phrases", AND / OR / NOT, prefix* and column filters.
- SemanticMultilingual bge-m3 embeddings find ideas, not just words.
- HybridBM25 and semantic results fused with Reciprocal Rank Fusion (RRF).
- Query fan-outAn LLM rewrites your query three ways; results are fused with RRF.
- JevOptional: when the top results are too close to call, TypeSafe’s Jev decides which one answers you.
- File over appJev pickstephango.combm25 · semantic
- Local-first softwareinkandswitch.comsemantic · fan-out
- The Markdown guidemarkdownguide.orgbm25
Illustration. Search is free — it never costs credits.
Connect once.
Every agent gets the context.
MCP over HTTP for Claude, Cursor and any MCP client, REST for apps, a CLI for pipelines, WebMCP for agents in the browser, and the URL prefix for a quick read.
MCP over HTTP
OAuth 2.1 + PKCE or API keys
CLI
Pipe Markdown anywhere
REST + OpenAPI
Interactive reference
WebMCP
Tools inside the page
// Claude Code
claude mcp add --transport http anymd https://anymd.cc/mcp
// Any MCP client (OAuth or API key)
{
"mcpServers": {
"anymd": {
"url": "https://anymd.cc/mcp",
"headers": { "Authorization": "Bearer amd_…" }
}
}
}
// Tools: read_url, search_library, get_document…https://anymd.cc/https://stephango.com/saw # or skip the scheme https://anymd.cc/stephango.com/saw
curl https://anymd.cc/api/v1/convert \
-H "Authorization: Bearer $ANYMD_API_KEY" \
-H "Content-Type: application/json" \
-d '{"url":"https://stephango.com/saw"}'npm i -g https://cdn.anymd.cc/cli/anymd-cli-latest.tgz anymd login anymd https://stephango.com/saw > saw.md anymd search "file over app" --mode hybrid
// On anymd.cc, in-page agents see these tools: navigator.modelContext // read_url, get_page_markdown, // search_library, list_documents (signed in)
const res = await fetch("https://anymd.cc/stephango.com/saw", {
headers: { Accept: "application/json" },
});
const { title, markdown, word_count } = await res.json();Reads what it’s asked to. Nothing more.
anymd is a reader for agents, not a crawler. Publishers stay in control, and the rules below are enforced in code on every channel.
Requested URLs only
anymd reads the URLs a user or their agent asks for. It does not crawl or discover pages on its own.
Public content only
Only pages that anyone can reach on the open web, without an account.
No bypassing access
No login, paywall or CAPTCHA bypass. Terms forbid users from trying, too.
Respects robots.txt
Checked before anymd fetches a page itself, for the anymd token or *. Disallowed pages are refused.
Domain opt-out
Site owners can block their domain on every channel, including cached results.
Per-site rate limits
Each site has a shared per-minute fetch budget, so agents never hammer a server.
Private library
What your agents read is saved to your account only. Never shared, never used to train models.
Takedown process
Abuse reports and takedown requests go to a human at hello@digitop.ai.
A dedicated reader
for each kind of source.
Each source gets its own extraction path, so a GitHub issue keeps its thread, a video keeps its transcript, and a spreadsheet stays a table.
Any web page
Articles, docs, blogs — clutter removed
X / Twitter
Posts, quotes, media, engagement stats
Public posts, captions and media; comments on request
Public posts, reels and media; comments on request
Threads
Public posts and carousel media; comments on request
Public posts, media and comments on request
YouTube
Title, channel and a timestamped transcript
GitHub
READMEs, issues, PRs and discussions
Threads with nested comments
Hacker News
Stories plus the top of the discussion
Text layout kept, tables preserved
DOCX · XLSX · CSV
Office files and sheets as Markdown tables
Images
Vision model describes and transcribes
Boring where it matters.
Proven extraction, open code, edge infrastructure and privacy defaults you do not have to ask for.
Proven engine
Battle-tested content extraction plus dedicated source adapters, all behind one conversion pipeline with per-request traces.
Open source, MIT
Read every line, open an issue, or self-host it on your own Cloudflare account.
Cloudflare edge
Workers, D1, Vectorize and R2 in 300+ cities. Results cached at the edge for an hour.
Private by default
Your library is yours. We never sell data or train models on what you convert. Delete anything, anytime.
Secure access
API keys stored as hashes, role-scoped permissions, OAuth 2.1 with PKCE for MCP clients.
Fair pricing
Pay for new sources, not seats. Cached reads and search are free. 14-day money-back.
Pay when your agents learn something new.
Reusing what they already know is free. Processing a new source uses credits by complexity — 1 for a web page — while cached reads and library search cost nothing. Pro and Scale never block you: overage is billed per 1,000 credits.
Free
For trying it on real work.
- 500 credits / month
- Library up to 1,000 docs
- BM25 + full-text + semantic search
- API, CLI, MCP & WebMCP
- 2 API keys
Pro
For builders and power users.
- 10,000 credits / month
- Unlimited library
- Query fan-out + Jev re-ranking
- Unlimited API keys & OAuth apps
- Overage $1 per 1k credits — never blocked
- Traces & 90-day usage logs
Scale
For teams and agent fleets.
- 100,000 credits / month
- Everything in Pro
- Overage $0.60 per 1k credits
- Priority conversion queue
- Role-scoped keys for your team
- Email support within 1 business day
Enterprise
Volume, SLAs, private deployments.
- Custom credit volume
- SLA & DPA
- SSO & audit exports
- Self-host on your Cloudflare account
Launch offer: 30% off Pro or Scale with LAUNCH30. Ends in .
Run your own context layer.
anymd is MIT-licensed and deploys as a single Cloudflare Worker with D1, KV, R2 and Vectorize. Keep your agents’ context on your own account.
Made by people who ship agents every day.
“Agents don’t need prettier websites. They need cleaner text.”
- ›CTO & Co-founder at TOPGROUP, DIGITOP & XINCHAO Live Music
- ›Founder of Build in Public VN
- ›Self-described "F*ck Around & Find Out" Specialist
- ›Ships open-source tools for AI agents: GoClaw, AgentKit, UI UX Pro Max
anymd plays well with friends.
Tools from Digitop.ai and nextlevelbuilder for building, running and remembering with AI agents.
Dewee
Multi-agent AI gateway built in Go. Single binary. Production-tested.
AgentBrain
AI data platform: semantic search and ETL your agents can query over MCP.
AgentKit
Complete AI engineering framework to build and review quality work.
GoClaw
Multi-agent gateway in Go with a lightweight desktop companion.
AgentWiki
A knowledge platform for humans and agents, with hybrid search and MCP.
UI UX Pro Max
Design intelligence skill for AI coding agents. 11K+ GitHub stars.
TOSE.sh
Git-to-Kubernetes PaaS: push code, get a running app.
Questions, answered.
Still curious? Email hello@digitop.ai — a human replies.
Is anymd free?
Yes. Anyone can prefix a URL with anymd.cc/ — no account, up to 50 reads a day. A free account gives you 500 credits a month plus a private, searchable context library. Paid plans start at $9/month.
What is a credit?
Credits pay for new source processing, priced by complexity: a web page is 1 credit, a YouTube transcript or a PDF/Office file is 3, an image is 5. Reusing what your agents already know is free: cached reads, library search and MCP recall never cost credits.
How is this different from fetching the HTML?
anymd is a source-aware reader: a proven content-extraction engine plus dedicated paths for sources such as GitHub, YouTube, Reddit, Hacker News, X and documents. Your agent gets the content, not the nav, cookie banners, ads or related posts, as structured Markdown with metadata, headings, links, tables, code and footnotes intact.
How do my agents use it?
Connect the MCP server at anymd.cc/mcp (OAuth or API key) and your agent gets read_url, search_library and get_document tools. Apps use the REST API, pipelines use the CLI, and in-browser agents get WebMCP tools on anymd.cc.
What happens to what my agents read?
Signed-in reads are saved to your private library so you and your agents can search them later. Nothing is shared, nothing is used to train models, and you can delete any document — or your whole account — at any time.
Does anymd crawl websites or get past paywalls?
No. anymd only reads URLs that a user or their agent asks for, and only content that is publicly reachable. It follows robots.txt, honours domain opt-outs, limits how often it hits each site, and never bypasses logins, paywalls or CAPTCHAs.
Is it open source?
Yes, MIT-licensed at github.com/digitopvn/anymd. You can self-host it on your own Cloudflare account.
What if I run out of credits?
On Free you wait for next month or upgrade. On Pro and Scale you are never blocked: extra usage is billed at $1 (Pro) or $0.60 (Scale) per 1,000 credits.
Can I get a refund?
Yes — 14-day money-back on your first subscription payment if you used less than 20% of the month’s credits. See the refund policy for details.
Give your agents the web as context.
Read your first source in five seconds. Keep everything in a private library your agents can search over MCP.
500 free credits every month · no card required