feat(mcp): register the maintain REST surface as local stdio tools

[luciferlive112116]

Summary

The maintain CLI subcommands call real REST endpoints that the remote server has exposed as tools since #6087, but the local stdio server registered none of them. An agent on the local server had to shell out to the CLI to list the approval queue, decide a staged action, toggle the kill-switch, set an autonomy level, or read gate precision.

This registers the five missing counterparts:

Tool	Endpoint	CLI equivalent
loopover_list_pending_actions	GET /agent/pending-actions	maintain queue
loopover_decide_pending_action	POST /agent/pending-actions/:id/:decision	maintain approve|reject
loopover_set_agent_paused	PUT /settings (agentPaused)	maintain pause|resume
loopover_set_action_autonomy	GET+PUT /settings (autonomy)	maintain set-level
loopover_get_gate_precision	GET /gate-precision	maintain precision

Each calls the endpoint its CLI subcommand already calls, through the same apiGet/apiPost/apiFetch client — so auth, timeouts and error shaping come from there rather than a second HTTP path, per the issue's "reuse the CLI's existing HTTP-client logic" requirement. Input shapes mirror the remote's, and categories mirror the remote's MCP_TOOL_CATEGORIES entries for the same names, so a caller sees one consistent surface across both servers. The remote server is untouched, as required.

One deliberate divergence from the remote — please read

The remote's listPendingActionsShape takes an optional status, which it can honour because it queries the approval-queue store directly. This server can only reach the queue through GET /v1/repos/:owner/:repo/agent/pending-actions, which takes no query parameters at all and hardcodes status: "pending" (src/api/routes.ts).

Mirroring the remote schema literally would therefore have shipped a filter that does nothing: a caller could ask for "rejected", receive the pending list, and be told it succeeded. That is precisely the failure mode of #6153 (merged in #6378), so I left status out of the schema and made the description name the queue it really returns. An agent picks its arguments from the published schema, so a filter that isn't there is one it won't ask for — and a key sent anyway is dropped by the MCP layer before the handler, so it can never reach the URL either. A test pins both halves.

set_action_autonomy read-merge-writes exactly as maintain set-level does: PUT /settings replaces the whole autonomy map, so sending one class alone would silently clear the others. A test asserts the other classes survive and that the write is a GET-then-PUT.

Validation

• New suite test/unit/mcp-cli-maintain-tools.test.ts — 15/15 pass: success + an API-failure path for each of the 5 tools, plus the read-merge-write invariant, the no-status-filter contract, and pre-flight rejection of unknown action classes / autonomy levels.
• mcp-tool-rename-aliases + mcp-cli-tools + mcp-cli-tools-search — 20/20 pass (the count canary, updated 42 → 47).
• 7 further mcp-cli/mcp-tool suites (basics, maintain, intake-tools, completion-spec, tool-categories, profiles, doctor) — 80/80 pass.
• npm run typecheck — 0 errors. npm run build:mcp — passes. npm run docs:drift-check — ok. git diff --check — clean.
• Rebased on latest main — no base conflict.

Coverage

No patch surface: coverage is collected over src/**, packages/loopover-engine/src/**, and packages/loopover-miner/lib/**. This PR touches only packages/loopover-mcp/** (not collected) and test/** (ignored). The new tools are still exercised end-to-end by the suite above, against the fixture server.

Scope

• One coherent change; wanted paths (packages/, test/). Remote server untouched; existing CLI subcommands' behaviour unchanged.
• No secrets/tokens/wallet/trust-score/reward terms; no changelog, site/, CNAME, or lovable changes.
• The alias-retirement suite intentionally pins the exact tool count — updated 42 → 47 with a note recording why, rather than loosened.

Safety

• These proxy maintainer-authenticated endpoints; the API enforces authorization exactly as it does for the CLI, and no tool decides access locally. Failure surfaces as a tool error rather than a silent empty success — asserted per tool.
• No secrets committed.

Closes #6152

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.60%. Comparing base (d280ce4) to head (aaa4392).
⚠️ Report is 25 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6382      +/-   ##
==========================================
- Coverage   95.60%   95.60%   -0.01%     
==========================================
  Files         599      599              
  Lines       47221    47238      +17     
  Branches    15024    15031       +7     
==========================================
+ Hits        45147    45160      +13     
- Misses       1290     1291       +1     
- Partials      784      787       +3     

Flag	Coverage Δ
shard-1	44.06% <ø> (-0.02%)	⬇️
shard-2	36.60% <ø> (-0.01%)	⬇️
shard-3	32.58% <ø> (+0.05%)	⬆️
shard-4	34.74% <ø> (+0.13%)	⬆️
shard-5	31.58% <ø> (-0.02%)	⬇️
shard-6	44.87% <ø> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

[loopover-orb]

Tip

✅ LoopOver review result - approve/merge recommended

_{Review updated: 2026-07-16 05:50:35 UTC}

3 files · 1 AI reviewer · no blockers · readiness 77/100 · CI green · clean

✅ Suggested Action - Approve/Merge

• safe to merge

Review summary
This PR registers five stdio MCP tools mirroring the remote server's already-shipped maintain REST endpoints (pending-actions list/decide, pause/resume, autonomy, gate-precision), reusing the existing apiGet/apiPost/apiFetch client rather than adding a new HTTP path. The set_action_autonomy tool correctly does a read-merge-write to avoid clobbering other action classes on PUT /settings, and the deliberate omission of a `status` filter on list_pending_actions is well-justified given the route hardcodes status=pending server-side. Tests cover registration, proxying, autonomy merge-preservation, and the intentionally-absent status filter, and the tool-count assertions in the alias-retirement suite were updated in lockstep (42→47).

Nits — 5 non-blocking

• packages/loopover-mcp/bin/loopover-mcp.js: the large block comment above the tool shapes duplicates rationale already in the PR description; consider trimming to keep the file size down given it's flagged as long (1603 lines).
• test/unit/mcp-cli-maintain-tools.test.ts: the 404 error-path test only checks the message matches /404|not_found/ rather than asserting on tool name specifically, so a shared fixture regression across all five tools would still pass individually — minor test-specificity nit.
• packages/loopover-mcp/bin/loopover-mcp.js: `decision` uses accept|reject (route vocabulary) while the CLI subcommands use approve|reject — this divergence is explained in a comment but could trip up a caller expecting CLI-consistent naming; consider surfacing this explicitly in the tool description too, not just the code comment.
• Consider extracting the five new tool registrations plus their shapes into a separate module given the file is already flagged as oversized (1603 lines, threshold 400), to ease future review diffs in this file.
• In gatePrecisionShape, confirm `windowDays` upper-bound isn't needed (e.g., a caller passing an extremely large integer) — currently only `.int().positive()` is enforced, mirroring the route's own lack of a cap, which is fine but worth a one-line note.

Decision drivers

• ✅ Code review — No blockers (1 reviewer)
• ✅ Gate result — Passing (No configured blocker found.)

Context & advisory signals — never blocks the verdict

Signal	Result	Evidence
Linked issue	✅ Linked	#6152
Related work	⚠️ 3 scoped overlaps	Top overlaps are listed below; lower-confidence bulk is hidden.
Change scope	❌ 8/20	High review scope from cached public metadata (1 linked issue).
Validation posture	✅ 25/25	PR body includes validation/test evidence.
Contributor workload	✅ 10/10	Author activity: 143 registered-repo PR(s), 81 merged, 32 issue(s).
Contributor context	✅ Confirmed Gittensor contributor	luciferlive112116; Gittensor profile; 143 PR(s), 32 issue(s).
Improvement	✅ Minor	risk: clean · value: minor · LLM: moderate

Linked issue satisfaction

Addressed
The PR registers all five requested tools (loopover_list_pending_actions, loopover_decide_pending_action, loopover_set_agent_paused, loopover_set_action_autonomy, loopover_get_gate_precision) in loopover-mcp.js, each backed by the same REST endpoints via the existing apiGet/apiPost/apiFetch client, and leaves the remote server and CLI subcommands untouched as required.

Review context

• Author: luciferlive112116
• Role context: outside_contributor
• Public audience mode: oss maintainer
• Lane context: Repository is configured for direct PR review.
• Public profile languages: not available
• Official Gittensor activity: 143 PR(s), 32 issue(s).
• Related work: Titles/paths share 6 meaningful terms. (PR #6383)
• Related work: Items reference the same linked issue feat(mcp): register the 8 write-tools (open_pr/file_issue/apply_labels/etc.) that packages/loopover-mcp's own recommendedTools already promise #6149. (issue #6149, PR #6383)
• Related work: Open PR work references issue feat(mcp): register the 8 write-tools (open_pr/file_issue/apply_labels/etc.) that packages/loopover-mcp's own recommendedTools already promise #6149. (issue #6149, PR #6383)

Contributor next steps

• Start here: Review top overlaps.
• Then work through the remaining 3 steps in the Signals table above.

Signal definitions

• Related work = same linked issue, overlapping active PRs, or title/path similarity.
• Change scope = cached public metadata such as size labels, draft state, and review-burden hints.
• Validation posture = whether the PR provides enough public validation/test evidence for maintainer review.
• Contributor workload = public contributor activity and cleanup pressure, not a repo-wide quality failure.
• Contributor context = public GitHub/Gittensor identity context; non-Gittensor status is not a blocker.

🧪 Chat with LoopOver

Ask LoopOver a question about this PR directly in a comment — grounded only in the same cached, public-safe facts shown above, never a new claim.

• @loopover ask <question> answers contribution-quality Q&A with source citations and freshness.
• @loopover chat <question> answers in natural prose from cached decision-pack facts via local inference (maintainer/collaborator; read-only).
• A plain-language @loopover mention with a real question is routed to the closest matching read-only command automatically — no exact syntax required.

Full command reference: https://loopover.ai/docs/loopover-commands

🧪 Experimental — new and may change.

_{🟩 Safe / merged · 🟦 Advisory · 🟨 Held for review · 🟥 Blocked / closed}

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by LoopOver, a quiet PR intelligence layer for OSS maintainers.

• Re-run LoopOver review

[loopover-orb]

LoopOver approves — the gate is satisfied and CI is green.