Pod Researcher MCP server

@pipeworx/pod-researcher

Podcast prospecting for guest booking and sponsorship: given a topic and a purpose, returns active podcasts whose recent episodes cover the topic, each with the best contact route the show itself published, the episodes that prove the fit, and your own outreach history so you never pitch the same show twice.

Part of Pipeworx — an MCP gateway connecting AI agents to 1681+ live data sources.

Tools

• findpodcastcontacts(topic, purpose, requireroute?, language?, country?, activeonly?, includeengaged?, limit?, cursor?) — active shows ranked by topical fit, each with bestroute (or routestatus: "nonefound" plus what was checked), evidence episodes, coverage counts and dataasof. purpose is guest or sponsor; routes designated for that purpose rank first, and a general contact or a person's published address is a labelled fallback. require_route: true returns only shows that have one.
• getpodcastprofile(id) — one show in full: identity, activity status, feed health, latest episodes, every published route. id is a show_id, an RSS feed URL (moved feeds resolve by their old URL) or a podcast:guid.
• refresh_podcast(id) — re-read one show's RSS feed now rather than waiting for its next scheduled check.
• watchpodcasttopic(topic, purpose, ...) / getpodcasttopicchanges(watchid?) — save a search and get the shows that newly match it.
• recordpodcastoutreach(showid, outcome, scope?, contactid?) — record contacted / replied / declined / booked / bounced / donotcontact / clear. Declined and donotcontact shows leave your searches; a scope: "person" record hides every show that publishes that person's address. Private to your account.

Auth

Keyless to search. watchpodcasttopic, getpodcasttopicchanges and recordpodcast_outreach need a signed-in account (an API key from pipeworx.io/signup): outreach history is kept per account, and an anonymous caller has none. This pack runs only on the Pipeworx gateway; a standalone build refuses every call rather than returning an empty list.

Contacts: what is returned and how it is labelled

Every address a show publishes in its own RSS feed or on its own website is returned, labelled:

• holdertype: role (a function inbox such as booking@ or sponsorships@), hostingaccount_owner (the itunes:owner address from the feed, usually the host), or person (an individual's own address).
• purpose (guest, sponsor, general) and confidence (designated when the show invited that purpose in words, fallback otherwise), plus match on the best route: designatedforpurpose or general_fallback.
• sourceurl (the page or feed it was read from) and lastobserved.

Addresses are never synthesised. A global removal list is honoured at serve time. recordpodcastoutreach never takes an address: a call carrying anything email-shaped is refused without repeating it, and a person-level record stores a SHA-256 of the address, not the address.

Data sources

• Podcast Index open directory () — which podcasts exist, their feed URLs and identifiers.
• Each show's RSS feed — episodes, publishing cadence, the itunes:owner address and invitation lines in episode notes.
• Each show's own website — its guest, advertise and contact pages.

Matching is on words in episode titles and show notes (the latest ~30 episodes per show), so a show that never names the topic in its notes is not found. Generated-content spam is left out by a classifier that also catches some real shows whose episode notes disclose AI assistance. Contact coverage is still growing: route_check on each result says whether the show's website has been read yet, so "none found" and "not looked at yet" are distinguishable.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

Shortened. The full README is on GitHub.