AI assistant

Connect the MCP server and an assistant can do most of what the other recipes describe, in conversation, without writing code.

Set up

1. Create an API key. For a first run, keep the default full scope set; narrow it later.
2. Connect a client. For Claude, that is one custom connector pointing at https://mcp.postal.music/mcp.
3. Ask: "How many tracks do I have, and what's my dominant genre?"

What works well

• Questions about the catalogue. Counts, filters by tag and date, who is on a track, where it was submitted.
• Organising. Building and reordering playlists, renaming, fixing metadata in bulk.
• Tagging. Triggering analysis, reviewing suggestions, applying or correcting them.
• Finding music for a brief. search_tracks with hard constraints, then a playlist link.
• Weekly reporting. Dashboard and top tracks, summarised in plain language.

See Example prompts for phrasing that gets good results.

What to be careful with

• update_track_tags replaces categories and trains the model. Ask the assistant to show the change before applying it.
• delete_composer is permanent.
• search_tracks costs a credit per call. Ask for a name or tag lookup when you already know what you want.
• Give a read-only key to assistants that only need to answer questions. See Suggested key profiles.

Building your own agent

The MCP server works with any client library that speaks Streamable HTTP. Fetch the live tool catalog from https://mcp.postal.music/tools to get JSON schemas for every tool, and authenticate with X-API-Key. Compact responses are the default; pass detail: "full" when your agent needs the raw object. See Compact mode.