Unit 3: AI-Agent Integration
Time: ~10 minutes • Lessons: 1 • Previous: Unit 1 — CLI track or Embedded track
Wire the workbench into a language model over MCP. The agent gains a bowire.discover / bowire.invoke / bowire.recordings.list toolset and can drive any of your APIs in plain English.
Prerequisites
- Unit 1 complete on either track (CLI or Embedded) — understand the discovery + invoke surface before handing it to an agent.
- Claude Desktop or Cursor installed locally. Any other MCP-aware host (custom MCP gateway, Continue.dev, &c) works too — the config snippet is portable; only the location of the
mcpServersconfig file changes. - The Unit 1 sample servers (
HelloApi+HelloGrpc) ideally still running — the lesson drives the agent against them.
Lessons
| Lesson | Topic | What You'll Build |
|---|---|---|
| 3.1 | Claude Desktop + Cursor over MCP | bowire mcp serve wired into the agent's mcpServers config, drive REST + gRPC + recordings from the chat window |
Why this unit
Bowire and MCP intersect in four distinct ways; Unit 3 covers the one that lets an agent drive Bowire. Same SDK, same install, same workbench — but the consumer is a Claude or Cursor chat instead of a browser. Useful for: "explore an unfamiliar API by asking questions about it", "regression-check that endpoints A through Z still return the right shape", "spin up a mock fixture for a downstream test from natural-language instructions".
Next: → Unit 4: Extending Bowire