/waveassist

Agents, built over MCP.

You do not write WaveAssist agents by hand. Your coding agent does, over MCP. This section is how that works, end to end.

Overview

WaveAssist connects to your editor over MCP. You describe a recurring job in plain English; your coding agent (Claude Code, Cursor, any MCP host) designs the pipeline, deploys it to WaveAssist, runs it once on real infrastructure, and arms the schedule only after that run passes.

The model does the thinking once, at build time. After that the agent runs as codified Python, with the model contained to wrapped, schema-locked calls. Same structure every run. The platform page covers the determinism model in depth.

Connect & keys

Auth is a rotatable mcp_token, separate from your UID. Get it from the Connect MCP panel at app.waveassist.io, then hand it to your editor as a bearer token. Regenerate it anytime without touching your identity or other keys.

urlhttps://mcp.waveassist.ai/mcp
headerAuthorization: Bearer YOUR_TOKEN
Full connect guide, with copy-paste snippets

Building an agent

Once connected, you build by asking. Name the sources, the job, and the schedule in one request. Your coding agent handles the wiring, writes the pipeline against the SDK, and deploys it.

It then runs a verify pass on live infrastructure. If it passes, the schedule arms. If it fails, it holds, so a broken agent never goes live unattended.

Examples

Each of these is a single request. The agent designs, tests, and schedules the rest.

Credits

Every account starts with $2 of runtime credit. After that, credits are prepaid and pay-as-you-go, no tiers and no subscription. A run consumes a fraction of a cent: the model tokens for any wrapped calls, plus the runtime.

You add credits from the same dashboard as your token. Balance and usage live on the WaveAssist Credits page.