feat(mcp): add tools search <query> for fuzzy tool discovery

[nghetienhiep]

What

Add loopover-mcp tools search <query> (with --json) so a user or agent can find the right tool among
the combined local + remote surface without reading the full tools listing top to bottom or already
knowing the exact name. The query fuzzy-matches against each registered tool's name AND description, so
a term that only appears in a description still surfaces the tool.

Why

tools/tools --json only dump every locally-registered stdio tool in registration order — the only way
to find one is to scroll the whole list or know its exact name (confirmed in #6300; no search/filter
existed). Once the local and remote sets are unified into ~150 tools that no longer scales.

How

• Turned tools into a small dispatcher: tools search routes to a new toolsSearchCommand, and the
  bare tools / tools --json listing behaves byte-for-byte as before (its option parsing just moved
  inside the command).
• searchTools ranks matches best-first: a substring hit on the name beats a substring hit on the
  description, which beats a typo-tolerant hit; non-matches are dropped and ties break alphabetically for a
  stable listing.
• Typo tolerance reuses the CLI's existing levenshteinDistance (the same utility behind
  suggestCommand's "did you mean") instead of adding a fuzzy-match dependency, per the issue's guidance.
  The edit-distance budget scales with the query length.
• --json returns { query, count, tools }; the human view prints aligned name + description rows, or a
  friendly No tools match "<query>". line when nothing matches. Running tools search with no query
  prints a usage error.
• Declared search in CLI_COMMAND_SPEC so tab-completion (all four shells) and the typo-suggester pick
  it up, and documented it in --help. Kept it local-CLI-only — the remote server's tool list stays a
  separate concern, as the issue scopes it.

Testing

• New test/unit/mcp-cli-tools-search.test.ts drives the real CLI and covers a name match (ranked first),
  a description-only match (query absent from every tool name), a Levenshtein typo match, a no-match query
  (empty --json result set and the plain "No tools match" line), the human row output, and the
  missing-query usage error.
• npm run typecheck, npm run build:mcp, npm run test:mcp-pack, and the existing
  test/unit/mcp-cli-tools.test.ts + test/unit/mcp-cli-completion-spec.test.ts suites — all green.

Closes #6300

[superagent-security]

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

[loopover-orb]

Tip

✅ LoopOver review result - approve/merge recommended

_{Review updated: 2026-07-16 01:42:01 UTC}

2 files · 1 AI reviewer · no blockers · readiness 88/100 · CI green · clean

✅ Suggested Action - Approve/Merge

• safe to merge

Review summary
The AI review returned non-blocking notes for this change but did not include a separate narrative summary. Review the nits below before deciding this PR.

Nits — 5 non-blocking

• mcp-cli-tools-search test and the two doc-comments above `toolsSearchCommand`/`searchTools` restate the same rationale three times in the diff; consider trimming to one comment site.
• packages/loopover-mcp/bin/loopover-mcp.js scoreToolMatch: `budget = Math.max(1, floor(needle.length/4))` allows an edit distance of 1 even for 1-3 character queries, which can surface noisy unrelated matches for very short searches — consider requiring budget 0 below some minimum query length.
• toolsSearchCommand derives `query` via `args.find(arg => !arg.startsWith('--'))`, so a query that itself begins with `--` (e.g. searching for a literal flag-like term) can never be matched — worth a one-line note if that's an accepted limitation.
• Add a test where the query itself starts with `--` to document the current usage-error behavior for that edge case.
• Consider exposing the search's score/rank in `--json` output for debugging/tuning rather than only exposing the final ordering.

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	#6300
Related work	✅ No active overlap found	No same-issue or scoped active PR overlap found.
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: 112 registered-repo PR(s), 64 merged, 0 issue(s).
Contributor context	✅ Confirmed Gittensor contributor	nghetienhiep; Gittensor profile; 112 PR(s), 0 issue(s).
Improvement	✅ Minor	risk: clean · value: minor

Linked issue satisfaction

Addressed
The PR implements `tools search <query>` with `--json` support, reuses the existing levenshteinDistance utility instead of adding a dependency, matches against both name and description, and includes tests for name-match, description-match, typo-tolerance, and no-match cases as required.

Review context

• Author: nghetienhiep
• Role context: outside_contributor
• Public audience mode: oss maintainer
• Lane context: Repository is configured for direct PR review.
• Public profile languages: Python, Vue, C#, JavaScript, C, C++, Go, Objective-C
• Official Gittensor activity: 112 PR(s), 0 issue(s).
• PR-specific overlap: none found.

Contributor next steps

• Start here: Add a concise scope and risk note.
• Then work through the remaining 1 step 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://gittensory.aethereal.dev/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.