Expand MCP tool surface: is_safe, score_batch, explain_risk, health_check#7
Conversation
Add is_safe_to_transact, score_batch, explain_risk, and health_check MCP tools. Refactor shared scoring logic into _score_one helper. Add amount_usd parameter to score_counterparty. Include FLAG_DESCRIPTIONS constant with 17 human-readable risk flag explanations. Bump version to 0.4.0. New tools: - is_safe_to_transact: quick yes/no wallet safety check with threshold - score_batch: parallel scoring of up to 10 wallets with semaphore(5) - explain_risk: human-readable flag explanations with tier recommendations - health_check: API availability check Tests: 45 new tests across test_mcp_tools.py and test_explain_flags.py. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ff9d59e721
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if amount_usd is not None: | ||
| body["amount"] = amount_usd | ||
|
|
||
| if not body: | ||
| return {"error": "At least one input field is required (domain, ip, wallet_address, stellar_wallet, or company_name)"} |
There was a problem hiding this comment.
score_counterparty now adds amount_usd to body before checking whether any identifying field was provided, so a call like score_counterparty(amount_usd=100) bypasses the documented “at least one input field” guard and proceeds to _score_one with no counterparty identifier. This violates the tool contract and can trigger avoidable failed (or billable) upstream requests; the emptiness check should be based only on identifier fields (domain, ip, wallet_address, stellar_wallet, company_name).
Useful? React with 👍 / 👎.
| # Sort by score descending | ||
| results.sort(key=lambda r: r.get("score", 0), reverse=True) |
There was a problem hiding this comment.
score_batch currently sorts with reverse=True, which places highest scores first, but the MCP tool instructions describe this output as “sorted by risk (lowest score first).” Consumers that use the first element as the riskiest counterparty will get the opposite ordering, so this should either sort ascending by score or have the contract text updated to match behavior.
Useful? React with 👍 / 👎.
- R1-2: Add exception handling to _call_direct (network errors no longer crash tools) - R1-3: Add domain/ip/company_name validation to explain_risk (matches score_counterparty) - R1-5: Fix sort order description in MCP instructions string Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
[Claude QA Review] Summary7 findings total: 3 auto-fixed, 2 need manual review, 2 informational (below threshold) Auto-Fixed (commit 040e528)
Needs Attention
Plan Coverage
All 9 plan items: IMPLEMENTED. No scope creep detected. Review by Claude Code — 3 review agents (code quality, plan coverage, integration) + auto-fix cycle |
|
[Claude CI Monitor] CI completed with failures: test (3.12), test (3.10), test (3.11) |
…lity The 127.0.0.1.evil.com hostname validation vulnerability was already fixed — the xfail(strict=True) markers caused XPASS failures in CI. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
[Claude CI Monitor] All CI checks passed on re-run (after xfail fix). |
- R1-7: Add score_batch partial failure test (2 successes + 1 error, verifies both results and errors buckets populated correctly) - R3-10: Extend defensive docstring test to cover all 4 tools that accept user identifiers (score_counterparty, is_safe_to_transact, score_batch, explain_risk). health_check exempt — no user inputs. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2 participants
Summary
Expands Revettr MCP server from 1 tool to 5 tools, making the scoring surface integration-ready for Bankr, Gina, and Robonet partnerships. Extracts shared
_score_onehelper, addsamount_usdto existing tool, and creates 4 new tools with full test coverage.Changes
_score_one(body)shared helper for payment path routingscore_counterpartynow acceptsamount_usdparameteris_safe_to_transact— binary safety gate with configurable thresholdscore_batch— parallel scoring of up to 10 wallets with Semaphore(5)explain_risk— human-readable risk narrative via FLAG_DESCRIPTIONS mapping (17 flags)health_check— API liveness probePlan Reference
Plan file:
~/.claude/plans/harmonic-sprouting-melody.md— Agent A workstreamTest Plan
pytest tests/ -v— all 45+ tests passpython -m revettr_mcplists all 5 toolsis_safe_to_transactreturns{"safe": true/false}with correct threshold logicscore_batchcaps at 10 wallets, sorts by score descexplain_riskmaps known flags to sentences, unknown flags get generic fallbackhealth_checkreturns API status on 200, error dict on failureGenerated with Claude Code