can I see if it is working?

[pvdhagen]

Stupid title, I know.

But as a matter of fact, I installed code-review-graph for opencode and than... opencode did everything as usual. And I as wondering "is code-review-graph used at all? Would I see in opencode some output "got information X from code-review-graph" or is it just a silent background service? Started asking Claude, which finally told me "CLAUDE.MD and AGENTS.MD tell me to use the mcp server, but it is not running, so I'm silently using fallbacks". Which is annoying.
I'm trying some propmting like "don't do a silent fallback, report issues". Can I activate some logging, like
mcp started
handling request
handling request
mcp shutdown
Report: "detect_changes" called 3 times, "refactor_tool" called once

?

I'm really worried it might fail silently again.

[phil-hudson]

+1 it's really not clear

[tirth8205]

You were right that this wasn't visible enough — the agent's own words ("it is not running, so I'm silently using fallbacks") mean the MCP server wasn't registered/started for that session, and nothing surfaced it.

How to verify today (v2.3.5):

1. code-review-graph status — graph stats + a staleness warning if it's out of date.
2. code-review-graph detect-changes --brief — prints a boxed Token Savings panel; that panel existing is proof the graph answered.
3. In Claude Code, type /mcp — code-review-graph should be listed as connected. If it isn't, docs/TROUBLESHOOTING.md item 5 has the exact checklist (.mcp.json present, command resolvable, restart the client).
4. The new docs/FAQ.md (PR feat: custom languages, PR-review GitHub Action, honest benchmarks, FAQ, community infra + verified Windows/hygiene fixes #541) has a dedicated "How do I verify it's working" section.

A doctor command that runs this whole checklist in one shot is on the roadmap — this thread is the reason.