absorb.md
JSON-RPC 2.0 · public · no auth

MCP endpoint.
Your AI agents
query absorb.md directly.

A live JSON-RPC 2.0 endpoint at https://absorb.md/api/mcp. Compatible with Claude Code, Cursor, Aider, Continue, and any MCP client. Six tools across 286 people, 29,346 compiled entries, and 30 topic wikis.


286
people tracked
29,346
compiled entries
30
topic wikis
7,680
recommendations indexed

Numbers refresh on page load · queried directly from production Postgres


Six tools. Plain JSON-RPC 2.0 over HTTP.

Every tool returns structured JSON with citations back to the original source URL. No scraping, no HTML parsing, no rate limits on the public endpoint.

list_people
Enumerate every tracked thinker with handles, bios, and entry counts.
Show me every AI thinker absorb.md tracks.
get_person
Fetch one person's full mega-wiki: bio, summary, content, themes, source breakdown.
What does Karpathy think about synthetic data right now?
search_entries
Keyword search across compiled entries. Optionally scope to one person.
Find all mentions of 'test-time compute' across every tracked thinker.
search_wikis
Keyword search across topic wiki articles.
What's the latest synthesis on LLM agent architectures?
get_wiki
Fetch a full wiki article by slug, with key claims and counter-claims.
Give me the full /wiki/quantum-computing article.
evaluate_against_projects
Returns a ready-to-run prompt template the caller's local LLM uses to score new entries against their active projects.
Score these new entries against my active projects. What should I ship this week?

Copy. Paste. Run. They work.

Three real curl calls against the production endpoint. No API key, no headers beyond content-type, no signup.

01 · tools/list
# Discover every tool the server exposes
curl -X POST https://absorb.md/api/mcp \
  -H "content-type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
02 · get_person karpathy
# Read Karpathy's full mega-wiki
curl -X POST https://absorb.md/api/mcp \
  -H "content-type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "get_person",
      "arguments": { "handle": "karpathy" }
    }
  }'
03 · search_entries 'test-time compute'
# Search every compiled entry for "test-time compute"
curl -X POST https://absorb.md/api/mcp \
  -H "content-type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "search_entries",
      "arguments": { "query": "test-time compute", "limit": 10 }
    }
  }'

One command. No API key. No signup.

Pick your tool. Run the command or paste the config. You're done.

Claude Code
run this in your terminal
claude mcp add absorb-md --transport http https://absorb.md/api/mcp
Cursor
paste into .cursor/mcp.json (project) or ~/.cursor/mcp.json (global)
{
  "mcpServers": {
    "absorb-md": {
      "url": "https://absorb.md/api/mcp"
    }
  }
}
Windsurf
paste into ~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "absorb-md": {
      "serverUrl": "https://absorb.md/api/mcp"
    }
  }
}
Claude Desktop
add via Settings > Connectors, or paste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "absorb-md": {
      "transport": "http",
      "url": "https://absorb.md/api/mcp"
    }
  }
}

Any MCP client that supports HTTP transport works. POST JSON-RPC 2.0 to https://absorb.md/api/mcp


Every AI newsletter is read-and-forget.
absorb.md is read-and-query.

Your agents ask structured questions and get cited claims with provenance back to the source URL. No other AI knowledge product ships a public MCP endpoint. The newsletters in your inbox cannot be queried by Claude Code; absorb.md can.


Public MCP endpoint is free. For private MCP with per-portfolio custom topic slices, see /for-vc for enterprise pricing.

Want to also run the local CLI? Connect in 30 seconds →