Prof

Local pprof & trace workbench for Go.
Pull profiles from a running service, keep them in one place, and explore them in the browser—or let Cursor read the same data via MCP.

Is this for you?

Yes, if you:

• Run Go services with net/http/pprof (or any reachable profile URL)
• Want a simple UI instead of juggling curl, files, and go tool pprof flags every time
• Need saved snapshots (heap, CPU, goroutine, trace, batch jobs) you can reopen later
• Use Cursor and want the agent to inspect profiles (top, source, flame, peek) on your machine

Probably not, if you only need a one-off go tool pprof on a single file and never collect from a live server.

What you get

• Collect — one-shot or batch fetch from http://your-app:6060/debug/pprof, or upload a .pprof / .trace file
• Browse — history of captures under local storage; open standard pprof views (graph, flame, top, source, diff)
• Ask AI (optional) — connect Cursor through MCP; the UI can generate your mcp.json with the right binary path and port

Requires graphviz installed if you use the graph view.

Try it in one minute

go install github.com/matteo-gz/prof/cmd/prof@latest
prof

Open http://127.0.0.1:8201, paste your app’s pprof base URL (e.g. http://127.0.0.1:6060/debug/pprof), pick a profile type, and submit.

From source:

cp env.yaml.example env.yaml
go build -o prof ./cmd/prof ./cmd/prof/wire_gen.go
./prof -c env.yaml

Cursor / MCP

1. Keep Prof running (web UI on port 8201 by default).
2. In the UI, click Use MCP → copy the suggested config, or see docs/mcp.md.

More

• Docker: prof_compose
• HTTP API & MCP tools: docs/mcp.md

License

Apache 2.0 — LICENSE