Connecting
One line, in whatever project you are working in:
claude mcp add --transport http archive https://democracyinnovators.com/api/mcpIt speaks the Model Context Protocol over HTTP, so anything that speaks MCP — Claude Code, Codex, your own client — connects the same way. No account, no key, nothing to install.
What it can do
Five tools appear once connected:
- search_transcripts — find passages by what was said, with the speaker, the second, and a link that opens the audio there.
- list_episodes — browse the archive, filtered by topic or guest.
- get_episode — one episode’s chapters, guests and publication date.
- get_transcript — the full transcript as speaker turns, or just the window between two timestamps.
- cite — a formatted reference for a moment: who was speaking, the episode, the year, the timestamp, and the link that proves it.
It returns sources, not answers
The archive hands over evidence; the assistant you connected does the writing. That is deliberate. If you are quoting these conversations in a thesis, an article or a report, you want the passage and a way to verify it — not a paraphrase standing between you and the recording.
Two limits worth knowing
Transcripts are generated automatically. They are good, and they are not perfect: names are the usual casualty. Every result carries a link to the moment it came from, and for anything you intend to quote, listen to it first.
Keyword search is free and unlimited. Searching by meaning rather than by wording costs a call to an embedding model from a small daily budget, so it needs a key. Ask us and we will issue one — for an agent that does its own reasoning, keyword search is usually enough.
Using the material
These are conversations people gave us in good faith. Quote them, study them, build on them — and credit the speaker and the episode when you do. If you are doing something substantial with the archive we would genuinely like to hear about it.