fix: add an MCP status command and report discovered-config drift

[sahrizvi]

Stacked on #1159. Base is fix/mcp-error-diagnostics, not main — this PR's diff is only the second commit. delete_branch_on_merge is off on this repo, so GitHub will not auto-retarget when #1159 merges. Someone has to run gh pr edit <this> --base main at that point.

1. No mcp status command (#790)

status is the name people reach for when a server will not connect, and it was the one name that did not exist.

Worth being straight about the size of this: the gap was much narrower than the issue implies. mcp listalready probed live — mcp.status() initialises the MCP service — and already printed the failure reason for failed and needs-client-registration servers. The missing piece was the entry point, not the view. So status is registered as a sibling sharing the same handler, rather than a second implementation to keep in sync.

Why a command and not an alias. It was first written as aliases: ["ls", "status"]. That widened yargs' alias column enough to rewrap an unrelated sibling row — mcp auth list lost its [aliases: ls] onto a second line. As its own command the help output stays additive: the diff against the committed snapshot is exactly one added line.

2. Discovery skips config drift (#878)

Discovery is first-source-wins, so a server already present in the user's config is skipped outright. When .vscode/mcp.json changes — a new ALTIMATE_EXTENSION_RPC port, a moved command — the difference was never mentioned. The reporter saw their datamate server keep a stale value with no indication why.

driftFields() reports which fields disagree, naming nested keys individually (environment.ALTIMATE_EXTENSION_RPC) so the message points at the thing to fix rather than just environment. enabled is excluded, since discovery sets it for its own reasons.

The configured value still wins. The command reports the disagreement and where to look, and leaves the decision to the user. Silently overwriting someone's own config would be worse than the silence it replaces — which is why this reports rather than reconciles, one of the three options the issue offered.

Testing

• Full opencode suite: 11,495+ pass, no regressions.
• e2e drives the real CLI against an isolated HOME plus a temp project containing a drifted .vscode/mcp.json, asserting both the drifted field is named and that an agreeing config stays silent.
• Mutation-tested: disabling drift recording, or the status registration, each fails exactly one test.
• Marker Guard passes; no new lint or formatting violations.

On flaky tests

The subprocess-heavy suites (test/pty, test/cli/run) fail intermittently under parallel load on this branch — different tests each run, all passing in isolation across repeated runs. These flakes pre-date this PR, but this PR does make them more likely to surface: it adds two more e2e files that each spawn real CLI subprocesses, increasing contention. Flagging it rather than leaving it to look like noise.

Test-harness fix worth knowing about

The e2e harness pattern copied from mcp-add.test.ts uses bun run --cwd <pkg>, which makes the CLI's working directory the repo package — so it read the repo's own .opencode config and never saw the temp project, meaning discovery never ran and the drift assertion silently could not fire. Fixed here by setting the spawn cwd to the project instead. mcp-add.test.ts is unaffected in practice because it passes --global, so it is left alone.

🤖 Generated with Claude Code

https://claude.ai/code/session_018fJ3X7pcGT4R9yzjsJnqsV

---

Summary by cubic

Adds mcp status so a server that will not connect has the command name people reach for. It reuses the mcp list handler rather than duplicating a view that already probes live and prints failure reasons; it's a distinct command instead of an alias because an alias widens yargs' help column and rewraps unrelated sibling rows.

Reports discovered-config drift. Discovery is first-source-wins, so a server already in altimate-code.json is skipped and a changed .vscode/mcp.json (new port, moved command) was never mentioned. mcp status and mcp list now name the differing fields — nested keys individually, e.g. environment.ALTIMATE_EXTENSION_RPC. The configured value still wins; enabled is excluded since discovery sets it for its own reasons.

Closes#790 and #878.

^{Written for commit 6776a8a. Summary will update on new commits.}

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
auto_review:
drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

👋 This PR was automatically closed by our quality checks.

Common reasons:

• New GitHub account with limited contribution history
• PR description doesn't meet our guidelines
• Contribution appears to be AI-generated without meaningful review

If you believe this was a mistake, please open an issue explaining your intended contribution and a maintainer will help you.