Skip to main content
TCW

DOCS

Technical documentation for the machine-readable layer.

⚡ MCP Server

The Context Window exposes a Model Context Protocol server that lets AI agents — Claude Code, Cursor, custom agents — query our knowledge graph directly. No scraping, no HTML parsing. Structured data over JSON-RPC.

Connection

ProtocolJSON-RPC 2.0 over HTTP POST
Endpoint/api/mcp
DiscoveryGET /api/mcp returns server info + tool list
AuthNone required (public, rate-limited)

Available Tools

search_articles(query: string, limit?: number)

Semantic search across all articles. Returns title, slug, TL;DR.

get_article(slug: string)

Full article by slug — key claims, predictions, metadata.

get_concept(name: string)

Look up an AI/tech concept — definition, related concepts.

list_topics((none))

All topic clusters with descriptions and article counts.

get_predictions(status?: string)

Timestamped predictions with confidence and tracking status.

Example

curl -X POST https://thecontextwindow.com/api/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "tools/call",
    "params": {
      "name": "search_articles",
      "arguments": { "query": "how do AI agents work" }
    },
    "id": 1
  }'

Claude Code Integration

claude mcp add tcw https://thecontextwindow.com/api/mcp

After adding, Claude Code can search articles, look up concepts, and query predictions directly.