The Wave Agents Toolkit: Connect Your Data to Any Other Tool
Three ways to get your Wave recordings into other tools — MCP for AI assistants, a REST API for your own code, and a CLI for the terminal.
Everything Wave captures — transcripts, summaries, action items, speaker labels, audio and video files — can be read from outside the Wave app. Wave Agents is the name for the three ways to do that: MCP for AI assistants, a REST API for code you write, and a command-line tool for scripts. All three run against your own account, use access you grant and can revoke, and are free on every Wave plan.
Pick the one that matches the tool you're connecting.
MCP — for AI assistants
The Model Context Protocol lets an AI assistant query your Wave library during a conversation. Claude, ChatGPT, Claude Code, Cursor, Windsurf, and Zed all speak it. Setup is a single URL — https://mcp.wave.co — and a browser sign-in. No code, no config files.
Once connected, the assistant can search your meetings by meaning, list recent sessions, open a summary, pull a speaker-labeled transcript, and read action items. It can also write action items back to a session and file sessions into folders. It cannot delete a session or rewrite a recording or transcript.
You can keep an assistant out of your personal recordings by scoping requests to a folder or a tag — "only use sessions in my Work folder" works, because the underlying tools accept folder and tag filters. See all about folders in Wave.
Setup guides: Claude, ChatGPT, and other AI tools.
Developer API — for code you write
A REST API at https://api.wave.co/v1, authenticated with a scoped token you create in Wave settings. Use it to sync sessions into a CRM, back up transcripts, build a dashboard, or feed a custom agent.
It covers listing and filtering sessions, summaries, transcripts, action items, semantic search, folders, stats, bulk export of up to 50 sessions in one request, and signed download URLs for audio and video. It also pushes webhooks to your server the moment a recording finishes processing, so nothing has to poll.
Start here: Wave Developer API.
CLI — for the terminal
@waveai/cli installs with npm and gives you a wave command for scripts, cron jobs, and one-off exports. Every command that returns data has a --json flag, so it composes with the rest of your shell.
Start here: Wave CLI.
What these tools can and can't do
Nothing is connected until you connect it, and every connection is scoped to your own account — no one else's recordings are ever reachable.
Permissions are not read-only by default. When you create an API token you tick exactly the permissions you want, and the list includes Write Sessions & Action Items, Delete Sessions, and Manage Folders. Signing the CLI in through your browser grants every permission, including deleting sessions. MCP is the narrow one: it can write action items and organize folders, but it has no delete.
Grant only the permissions an integration actually needs. A token with Delete Sessions can permanently remove recordings, and API deletes don't behave like the in-app trash — see Recently Deleted.
Revoke anything at any time from the Integrations tab in your Wave profile on the web, under Developer API.
Hand the docs to your AI assistant
Every page on api.wave.co has a copy button in the header — Copy integration docs, or Copy MCP docs on the MCP page. It copies a plain-text description of the whole platform — endpoints, scopes, payload shapes, examples — that you can paste into Claude or ChatGPT and say "write me the integration." The same text lives at api.wave.co/llms.txt, and the full interactive reference is at api.wave.co/reference.
Support
For setup help, use the in-app chat at app.wave.co or email support@wave.co. For questions about building on the API, MCP, or CLI, api.wave.co is the most complete and most current reference.
Was this article helpful?