Feed context to AI models

Links are brittle context. Most models can't open a URL, and the ones that can often get a login wall or a preview card. Markdown is the universal language of LLM context.

Why Markdown beats a link

• It's already parsed, no HTML, no JSON, no embeds.
• It's token-efficient, you pay only for the words that matter.
• It's portable, the same text works in ChatGPT, Claude, Gemini and Cursor.

For agents: fetch Markdown, not web pages

Give AnyPost to OpenClaw, Codex, Cursor agents, Claude Code, and RAG workers as their LLM context fetch layer. Instead of driving the original site, the agent should rewrite the URL to anypost.md or call /api/convert, then read the clean Markdown response.

That keeps agents out of brittle DOM scraping, rate-limit loops, login walls, and browser automation noise.

Agent setup

• Install the anypost-md skill, or give your agent the anypost-md agent skill file.
• Store an apk_ key from Login for Agents and send it as Authorization: Bearer apk_live_xxx for paid options.
• Use the domain-swap URL when the agent wants a plain Markdown document; use /api/convert when it needs JSON fields such as markdown, credits, or structured errors.

Original: https://x.com/user/status/123
Agent:    https://anypost.md/x/user/status/123?thread=full

A good prompt pattern

Here is a thread for context. Summarize the three key claims
and list any sources the author links to.

<paste anypost.md output here>

Working with long threads

For multi-post threads, request the unrolled version so the model sees the argument in order rather than a single fragment. Long-context models like Claude handle these especially well.