Claude can use tools now. Not metaphorically: since Anthropic shipped connectors, you can hand Claude a URL and it gains real abilities, with live data behind them. We run one of those tools, and this post is the shortest honest path to making Claude do your playlist research.
If you’d rather understand what you’re connecting first, read our plain-English explainer on what an MCP server is. If you just want it working, keep reading; the setup takes about a minute.
Step 1: add the connector
In Claude (web or desktop), open Settings → Connectors → Add custom connector. Name it PlaylistMap and paste:
https://playlistmap.com/mcp/server
That’s the entire setup. No API key, no account. Custom connectors require a paid Claude plan; if you’re on an older Claude Desktop build without the Connectors screen, there’s a bridge configuration (it needs Node.js installed) on our setup page.
Step 2: check the handshake
Ask Claude: “What PlaylistMap tools do you have?” It should report two: match_playlists and check_playlist. If it doesn’t, the connector didn’t attach; remove it and paste the URL again, or see the troubleshooting notes on the setup page.
Step 3: run a real research session
Here’s the sequence we recommend for one release, in plain prompts:
- “Here’s my single: [Spotify track link]. Which active playlists fit it? Show follower counts and why each one matched.” Claude calls the matcher against our catalog (70,000+ playlists from 37,000+ independent curator profiles) and returns a ranked list. Only currently-active playlists are considered; a playlist untouched for more than a year is unlikely to accept a new submission.
- “Rank the top ten by realistic value for an unknown artist.” Claude reasons over follower counts, fit strength and freshness. Expect it to favor mid-size playlists; a 3,000-follower playlist updated weekly is often a more practical pitch target than a dormant 40,000-follower one.
- “I’ve been offered a paid slot on this playlist: . Check it.” This is the integrity check: a 0–100 score from public signals like curator activity and engagement coherence. Scores under 55 mean look closer; under 35 means multiple risk patterns at once. Treat the result as screening, not proof.
- “Compile the shortlist with reasons, and give me the link to unlock the curators.” Claude hands back a signup link; contacts and the pitch itself happen inside PlaylistMap, not in the chat.
Optional: install the skill so Claude runs the whole flow itself
If you do this often, we publish a free Claude skill, “PlaylistMap Pitch Researcher”. It’s a single instructions file (no code, 1.6 KB, checksum published) that teaches Claude the complete workflow above: collect song details, match, vet, shortlist, hand off. Install it via Settings → Capabilities → Skills and then a single prompt like “research playlists for my new single” triggers the full sequence. Download and details are in the skill section of the MCP page.
What the connector cannot do
Two things, by design. It cannot read curator contact details, because the server never returns them to any agent. And it cannot send pitches, because there is no tool for that, on purpose. We wrote about the reasoning in more depth in our main AI-agent guide, but the short version: automated outreach is the fastest way to make curators stop reading pitches at all. Claude researches; you pitch.
Cost and limits, stated plainly
The research layer is free with per-hour rate limits (matching runs real analysis on our side, so the caps keep it sustainable). Unlocking curator contacts to actually pitch is the paid part of PlaylistMap, from a free account with starter credits. Nothing about the Claude connection itself costs anything beyond your Claude subscription.
One caveat: an agent makes research faster, not taste better. Claude can tell you “clairo vibes” fits your track and looks healthy. Whether your mix is ready for it is still your call.
