Machine access

Chain of Thought for AI agents

AI agents can search Chain of Thought episodes and transcript passages through the public MCP server at https://mcp.chainofthought.show/mcp. Each result gives a chainofthought.show link. Cite the episode page, or the linked moment when referring to a passage.

MCP server

The remote endpoint uses Streamable HTTP. It is stateless and read only. You do not need an account, login, or API key. The server code is available under MIT. Transcript text and episode metadata are licensed CC BY 4.0 and available as a dataset.

Available MCP tools
Tool	What it returns
`search_transcripts`	Full-text search over transcript passages, up to 20 results with a link to each moment.
`search_episodes`	Episodes matching a topic, guest, company, title, or chapter.
`get_episode`	Episode metadata, guests, chapters, and links. It does not return the transcript.
`get_transcript_excerpt`	Up to 16 consecutive transcript segments around a point.

Connect

Claude Code

claude mcp add --transport http chain-of-thought https://mcp.chainofthought.show/mcp

Claude on claude.ai or Desktop

Open Settings > Connectors > Add custom connector, then paste https://mcp.chainofthought.show/mcp.

ChatGPT

Where your plan supports custom connectors, add the remote MCP URL in developer mode connectors: https://mcp.chainofthought.show/mcp.

Cursor

Add this entry to mcp.json:

{
  "mcpServers": {
    "chain-of-thought": {
      "url": "https://mcp.chainofthought.show/mcp"
    }
  }
}

Other machine-readable endpoints

These public files come from the same episode archive as the HTML pages. Use the manifest to find current episode JSON URLs.

/llms.txt: an index of citable episode and reference pages.
/llms-full.txt: the episode map with takeaways and FAQ text.
/corpus.json: a manifest with episode metadata, counts, usage guidance, and each episode JSON URL.
/corpus.jsonl: newline-delimited transcript segments, each with a citation URL.
Open dataset (Hugging Face): transcript text and episode metadata as Parquet under CC BY 4.0.
/podcast/{slug}.json: one episode's metadata, chapters, and transcript. Follow the json field in the manifest for the current path.

How to cite

Cite the episode page URL supplied with a result. For a specific statement, use its moment link with the fragment anchor. The canonicalUrl field in corpus data points to the same HTML citation surface. Transcripts are machine-generated and can contain errors, so verify quotations against the recording before publishing them.

Limits and freshness

The MCP server limits each IP address to 10 requests per 10 seconds. Results can be cached for up to an hour. A new episode is usually searchable within about an hour after it appears on the site.

Privacy

The MCP server stores no accounts, cookies, or query history in its database. Cloudflare processes requests: IP addresses are used for rate limiting and appear in Cloudflare's operational logs, retained under Cloudflare's own policies. The Worker writes short operational log lines for debugging, including the tool name and error type, without query text. Tool results can be cached at Cloudflare's edge for up to an hour, keyed by the normalized request; the cache entry does not include your IP address.

Queries are not used to train models and are not sold or shared. For questions about this policy, contact Conor Bronsdon on LinkedIn.