API & MCP

Everything published in this guide can be read by an AI agent or a program: the same papers, news, researcher profiles and debate claims, with the same sources and AI-written labels. It is read-only, free, and needs no key or account.

Connect over MCP

The guide runs a Model Context Protocol server. Any MCP client can add it by URL:

https://api.aifaultline.com/mcp

Streamable HTTP transport, stateless, no sign-in.

Claude Code

claude mcp add --transport http ai-fault-line https://api.aifaultline.com/mcp

Claude (claude.ai and the desktop app)

Open Settings, then Connectors, choose Add custom connector, and paste the URL above. Leave the authentication fields empty.

Cursor, Windsurf and other clients that take a URL

{
  "mcpServers": {
    "ai-fault-line": {
      "url": "https://api.aifaultline.com/mcp"
    }
  }
}

VS Code

In .vscode/mcp.json:

{
  "servers": {
    "ai-fault-line": {
      "type": "http",
      "url": "https://api.aifaultline.com/mcp"
    }
  }
}

Clients that only run local servers

Bridge the URL with mcp-remote (needs Node.js):

{
  "mcpServers": {
    "ai-fault-line": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.aifaultline.com/mcp"
      ]
    }
  }
}

Tools

Tool	Arguments	Returns
`search`	query (required); types: paper, news, author, claim; limit 1–50	Full-text search over papers, news and podcasts, researcher profiles and the five claims.
`get_paper`	id (required): a paper key from search, or its site slug	Bibliographic data, abstract, AI-written summary and explainer when there is one.
`get_author`	id (required), e.g. daron-acemoglu	Role, documented-facts-only bio with sources, links, and documented positions with evidence.
`list_claims`	none	The five debate claims with how many coded researchers agree, are mixed or disagree.
`get_claim`	id (required): productivity, jobs, inequality, regulation or pace	The claim, its state-of-the-debate note with sources, and the coded positions.
`list_recent_news`	kind: article or podcast; topic; region (e.g. US, EU); limit 1–50	The newest published news articles and podcast episodes.
`get_letter`	none	The wemustactnow.ai letter: verbatim text and a labeled AI summary.

Results are JSON. A typical session searches first, then fetches the full record with get_paper, get_author or get_claim.

JSON API

The same content over plain HTTP GET, at https://api.aifaultline.com. There is an OpenAPI description, and CORS is open, so it works from a browser too.

Endpoint	Parameters	What it returns
`/api/search?q=…`	type (paper,news,author,claim), limit, theme, topic, region, kind	Search everything published
`/api/papers`	limit	Published papers, newest first
`/api/papers/{id}`	—	One paper by key or slug
`/api/news`	limit, kind, topic, region	Published news and podcast episodes
`/api/authors`	—	Every profiled researcher
`/api/authors/{id}`	—	One profile with documented positions
`/api/claims`	—	The five debate claims with stance counts
`/api/claims/{id}`	—	One claim with its note and coded positions
`/api/letter`	—	The open letter
`/api/status`	—	Record counts and the time of the last sync
`/export/`	—	The CSV and JSON downloads: files, record counts, sizes, build time

curl "https://api.aifaultline.com/api/search?q=productivity&type=paper,claim&limit=5"

Calling the MCP server by hand

Each request is one JSON-RPC message in a POST:

curl https://api.aifaultline.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"get_claim","arguments":{"id":"jobs"}}}'

Using what you get

Summaries, explainers, claim notes and researcher bios are AI-written; each record says so in its labels. Researcher positions are documented positions with citations, not characterizations. When you quote the guide, keep the source links, say where text is AI-written, and give stance counts with their denominator (“n of N coded researchers”). The scores used to curate the collection are internal and not exposed.

A plain-text overview for crawlers and language models is at llms.txt. Calls are logged anonymously to see which tools are used (usage statistics). Whole-collection CSV and JSON files are on Download.