Podcast Parser MCP server

@pipeworx/podcast-parser

What gets repeated across 162 tracked podcasts — claims extracted from 10,000+ episodes, each carrying the measurement behind its status: how many independent podcasts and speakers repeat it, excluding whoever originated it.

Part of Pipeworx — an MCP gateway connecting AI agents to 1576+ live data sources.

LIVE, BYO-ONLY. Promoted out of _incubator and wired into the gateway on 2026-09-14 (fleet #1958), once Bruce's partner key existed and all eight exported tools returned non-empty payloads against the real API.

It ships BYO-only even though a Pipeworx platform key exists, and that is a deliberate choice, not an oversight. The gateway is AT Cloudflare's limit of 250 text bindings, so wrangler secret put PLATFORMCANONCANNONKEY is rejected outright — "Too many text bindings, found a total of 251, they exceed the limit of 250" (code 10055). Declaring platformKeyEnv without the secret actually present would un-sink this pack in routing and send podcast-consensus questions to a tool that can only refuse, which is strictly worse than an honest nomatch — the fda-inspections precedent, fleet #617. So the pack declares no platformKeyEnv, keyBlockedTools() keeps it out of askpipeworx retrieval, and a caller with their own CanonCannon key can still call it directly with _apiKey.

To finish it: free one text binding on the gateway, then land the wrangler secret put and platformKeyEnv: 'PLATFORMCANONCANNONKEY' in the SAME change. The key itself is already issued, stored in the canonical checkout's gitignored .env, and verified live. canon_citations is separately held — see the integrity gate.

Tools

Eight are exported. canon_citations is implemented and deliberately held — see the integrity gate.

Each row carries tier and its measurement.

• canon_search(query?, tier?, label?, limit?, cursor?) — find recurring ideas.

books, plus the measurement spelled out in one sentence beside it.

• canon_get(slug) — the full entry: claim, evidence, nuance, gaps, related,

of the canon bar.

• canonepisodesearch(query?, podcast?, since?, limit?, cursor?) — the episode corpus.
• canonepisodeget(slug) — one episode with its extracted ideas, guests and books.
• canonpersonget(slug) — what one person has actually said, across shows.
• canonpodcastlist(limit?, cursor?) — the tracked-show roster, i.e. the denominator

and originator involvement.

• canonpodcastget(slug) — one tracked show.
• canoncorpusstats() — corpus counts and the bar currently in force.
• canoncitations(slug) — HELD.** Every episode citing an idea, with speakers

Auth

BYO only today: apiKey (Authorization: Bearer cclive...). Every /api/v1 endpoint refuses without one — GET /stats with no key is a clean 401 {"error":"Missing or malformed API key. Send Authorization: Bearer cclive_..."}, so a missing credential is unambiguous rather than looking like an outage.

There is no platformKeyEnv because we hold no key. When Bruce issues the partner key, wire PLATFORMCANONCANNONKEY on the gateway in the SAME change that adds the MCP_PACKS entry (fda-inspections precedent, fleet #617) — a declared-but-unset platform key sinks every tool in the pack under the key rule and reads to a caller as a broken pack rather than an unfunded one.

The citations endpoint is additionally restricted to an internal-tier key, so a partner key gets 403 there. That is upstream policy, not our gate.

The measurement envelope — the whole contract

tier is never served without a measurement sibling, in one of exactly two shapes:

{ "available": true, "measured_at": "…", "independent_podcasts": 23,
  "independent_speakers": 41, "independent_episodes": 36,
  "threshold": 9, "tracked_podcasts": 162 }

{ "available": false, "reason": "not_yet_measured" | "not_yet_persisted" }

Shortened. The full README is on GitHub.