Verdict: the due-diligence house for DeFi

verdict-finance-mcp

Verdict is the due-diligence house for DeFi: independent ratings, rating-priced cover, security testing and continuous monitoring, plus a security partner marketplace for audits, monitoring and pen testing, across seven entity types. This Model Context Protocol server lets AI agents call that house directly.

Live today: Ratings. Letter grade (AAA to D) plus a composite score (0 to 100) built from 300+ criteria across the full dependency graph, for protocols, chains, tokens, oracles, vaults, organisations and bridges. Plus post-quantum readiness for chains, and a live feed of DeFi hack incidents (confirmed by default).

Keyless by default. It wraps Verdict's live anonymous API, so there is no signup and no API key. Each user runs it locally and queries from their own IP. Bring your own key if you want the per-domain breakdown behind a grade (see below).

Install

Add to your MCP client config (Claude Desktop: claude_desktop_config.json; Claude Code: .mcp.json):

{  "mcpServers": {    "verdict-finance": {      "command": "npx",      "args": ["-y", "verdict-finance-mcp"]    }  }}

That is it. No key needed to read ratings. (Requires Node 20+.)

Tools

Every rating response ends with an attribution line linking back to the rating on verdict.finance.

Bring your own key

Ratings are free to read: the tools above work with no key at all. Set VERDICT_API_KEY and the server authenticates every call, which adds get_rating_breakdown (per-domain scores). Keys come from the Verdict dashboard at verdict.finance; plans are at verdict.finance/pricing.

{  "mcpServers": {    "verdict-finance": {      "command": "npx",      "args": ["-y", "verdict-finance-mcp"],      "env": { "VERDICT_API_KEY": "your-key" }    }  }}

Without the variable the server behaves exactly as it always has: no header is sent, and nothing about the keyless path changes. The key is never logged, echoed, or included in tool output. Called without a key, get_rating_breakdown answers with where to get one rather than failing.

Agent-native micropayment access (x402) is the announced direction. It is not shipped, and nothing here is behind a paywall today.

Quantum-readiness data is provided by LayerQu and is companion data only. It never feeds a Verdict grade; chain ratings simply carry an extra LayerQu line when a reading is available.

The incident feed carries hack incidents across DeFi (confirmed by default), including ones at protocols Verdict does not rate. Two things are worth knowing when reading it. Matching to a rated protocol happens when an incident is received, and re-runs when an incident escalates and its payload improves, so matched_protocol reads as of the incident's current tier. It is still never retroactive when Verdict's own coverage grows. And an incident is a signal to look, never an automatic downgrade: a matched entity may be flagged under review, but re-rating is always a human decision.

Example prompts

• "What does Verdict rate Aave?" runs get_rating({ entity_type: "protocol", identifier: "aave-v4" })
• "Search Verdict for Chainlink." runs search_ratings({ query: "chainlink" })
• "List the chains Verdict rates." runs list_ratings({ entity_type: "chain" })
• "Is Ethereum quantum-ready?" runs quantum_readiness({ chain: "ethereum" })
• "Any DeFi hacks this week?" runs get_recent_incidents({ limit: 10 })
• "How does Verdict score a bridge?" runs get_methodology({})
• "What dragged Aave's grade down?" runs get_rating_breakdown({ entity_type: "protocol", slug: "aave-v4" })
• "Get my protocol rated." runs request_coverage({ entity_name: "...", entity_type: "protocol", contact_email: "..." })

Example output

Aave V4 — BBB (78/100) · protocol · chains: Ethereum · categories: lending · https://www.verdict.finance/products/ratings#protocol-aave-v4
Rating by Verdict — https://www.verdict.finance/products/ratings#protocol-aave-v4 · Methodology + deeper analysis at https://www.verdict.finance

Entities that are not yet rated render as unrated. When a rating was pulled down by a weak dependency, the response says which one.

The roadmap: the callable marketplace

Verdict's diligence lifecycle is Assess, Rate, Cover, Secure, Monitor. Ratings are the first module agents can call. As the rest of the house becomes callable, this server grows a module per product:

• Cover: rating-priced parametric cover on rated protocols.
• Audit Marketplace: hand over a repo, receive blind bids from vetted audit firms.
• Monitoring Marketplace: watchlists, live alerts and webhooks on rated entities.
• Pen Test Marketplace: engage penetration tests through vetted partner firms, scoped and priced per engagement by the partner.

The rating function stays structurally separate from what the marketplace sells: buying a service never moves a grade.

Tool names are module-scoped (search_ratings, get_rating, list_ratings) so new modules arrive without breaking existing agents.

How it works

Verdict's read API is anonymous and rate-limited per IP (120/min, 20k/day). This server:

• lists and looks up entities via GET /{type}s and GET /{type}s/{identifier},
• fetches each entity's latest published scorecard (GET /{type}s/{id}/scorecards) for the grade + composite score,
• reads the published methodology from GET /methodology,
• reads per-domain scores from GET /{type}s/{slug}/domains (the one route that needs a key), and
• submits coverage requests to POST /contacts, the same public endpoint the "Get rated" form on verdict.finance posts to.

The API base is https://api.verdict.finance/api/v1 and can be overridden with the VERDICT_API_BASE environment variable (used for testing). Set VERDICT_API_KEY to authenticate; leave it unset to stay keyless. The full API surface is documented at api.verdict.finance/openapi.json.

Rate limits and outages are handled gracefully. Tools return a friendly text message, never a crash.

Attribution and trademark

Ratings and methodology are © Verdict. This tool surfaces the public ratings and the published methodology; the deeper analysis lives at verdict.finance. The code is Apache-2.0; the Verdict name and branding are not licensed by it.

Development

npm installnpm run build     # tsc -> dist/npm test          # vitest (unit tests, HTTP client mocked)
# optional live smoke against the real API:VERDICT_LIVE_SMOKE=1 npx vitest run test/live.smoke.test.ts
# the smoke's one keyed case (get_rating_breakdown) skips unless a key is set:VERDICT_LIVE_SMOKE=1 VERDICT_API_KEY=... npx vitest run test/live.smoke.test.ts

License

Apache-2.0. Copyright 2026 Verdict Capital.