AgentTrust
Challenge-response quality verification for AI agents and MCP servers.
AgentTrust
Challenge-response quality verification for AI agents and MCP servers.
AgentTrust evaluates AI agent competency before you trust them with real tasks or payments. It connects to any MCP server, runs challenge-response tests across 6 quality dimensions, and issues W3C Verifiable Credentials as proof.
Why
The AI agent ecosystem has identity (ERC-8004, SATI), post-hoc reputation (TARS, Amiko), and payments (x402) — but no pre-payment quality gate. AgentTrust fills this gap: verify competency first, then trust.
Features
Evaluation Engine
- 3-level pipeline: Manifest (schema) → Functional (tool calls) → Domain Expert (calibrated questions)
- 6-axis scoring: accuracy (35%), safety (20%), reliability (15%), process quality (10%), latency (10%), schema quality (10%)
- Consensus judging: 2-3 LLM judges in parallel with agreement threshold (saves 50-66% LLM calls)
- 7 LLM provider fallback chain: Cerebras → Groq → OpenRouter → Gemini → Mistral → DeepSeek → OpenAI
- 5 adversarial probe types: prompt injection, PII leakage, hallucination, overflow, system prompt extraction
Battle Arena
- Head-to-head blind evaluation with position-swap consistency
- OpenSkill (Bayesian ELO) rating system with divisions (Bronze → Grandmaster)
- Fair matchmaking: rating proximity + uncertainty bonus + cross-division challenges
- Style control penalties to prevent gaming via verbose/formatted responses
IRT Adaptive Testing
- Rasch 1PL calibration from battle data (pure Python, no numpy)
- Fisher information maximization for adaptive question selection
- EAP ability estimation with standard normal prior
- Reduces evaluation cost by 50-90% while maintaining accuracy
Standards
- W3C Verifiable Credentials (AQVC format) with Ed25519 DataIntegrityProof
- Google A2A v0.3 native support (AgentTrust IS an A2A agent)
- x402 Solana payment verification (USDC + SOL)
- AIUC-1 protocol mapping
Quick Start
Docker (recommended)
cp .env.example .env
# Add at least one LLM key (GROQ_API_KEY, CEREBRAS_API_KEY, etc.)
docker compose up -d
Services:
- API: http://localhost:8002
- MCP Server: http://localhost:8003
- Health: http://localhost:8002/health
Local Development
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
# Add LLM keys to .env
unset GROQ_API_KEY # Shell env overrides .env rotation pool
python -m uvicorn src.main:app --host 0.0.0.0 --port 8002 --reload
MCP Server (for Claude, Cursor, Windsurf)
Add to your MCP client config:
{
"mcpServers": {
"agenttrust": {
"command": "python",
"args": ["-m", "src.standards.mcp_server"],
"env": {
"GROQ_API_KEY": "your-key"
}
}
}
}
Or connect to a running instance via SSE:
http://localhost:8003/sse
Available MCP tools:
| Tool | Description |
|---|---|
check_quality(server_url) | Full evaluation: manifest + functional + judge scoring |
check_quality_fast(server_url) | Cached score (<10ms) or manifest-only (<100ms) |
get_score(server_url) | Lookup cached score with freshness decay |
verify_attestation(attestation_jwt) | Verify AQVC JWT and decode payload |
API Endpoints
| Method | Endpoint | Description |
|---|---|---|
| POST | /v1/evaluate | Submit target for evaluation |
| GET | /v1/evaluate/{id} | Poll evaluation status |
| GET | /v1/score/{target_id} | Get quality score |
| GET | /v1/scores | Search/list scores |
| GET | /v1/badge/{target_id}.svg | SVG quality badge |
| GET | /v1/attestation/{id} | Get signed attestation (JWT or W3C VC) |
| POST | /v1/attestation/{id}/verify | Verify attestation |
| POST | /v1/feedback | Submit production feedback (anti-sandbagging) |
| POST | /v1/battles | Create evaluation battle |
| GET | /v1/arena/leaderboard | Battle arena leaderboard |
| GET | /v1/rankings | Global rankings by domain/tier |
| POST | /v1/irt/calibrate | Trigger IRT batch calibration |
| GET | /v1/irt/recommend | Adaptive question selection |
| GET | /v1/pricing | x402 pricing table |
| GET | /.well-known/agent.json | A2A Agent Card |
Architecture
src/
api/v1/ # 14 FastAPI routers
core/ # Evaluator, MCP client, scoring, IRT, battle arena
auth/ # API keys (SHA256 + salt), rate limiting by tier
storage/ # MongoDB (Motor) + Redis
payments/ # x402 protocol, Solana verification
standards/ # W3C VC issuer, A2A extension, MCP server, AIUC-1
Stack: FastAPI + MongoDB + Redis | 533 tests | 60 source files | 15 lean dependencies
Tests
python -m pytest tests/ -q
# 533 passed in ~2s
Configuration
See .env.example for all 60+ configuration options including:
- LLM API keys (7 providers, comma-separated for rotation)
- MongoDB/Redis connection
- JWT attestation (Ed25519 key, issuer DID, validity)
- Solana wallet for x402 payments
- Rate limit tiers and consensus judge settings
License
MIT
Links
- Architecture — Full system design (845 lines)
- Distribution Roadmap — Partner and integration plan
- A2A Agent Card — Machine-readable capabilities
Built by Assisterr
相關伺服器
AsusWRT MCP Server
Model Context Protocol server for secure AsusWRT router administration via SSH. Provides 42+ read-only monitoring tools and guarded mutation tools for managing AsusWRT/Merlin routers.
Janee API Security
MCP server that sits between AI agents and APIs. Agents request access, Janee makes the call with the real credentials, agents never see the secrets.
MCP Goodnews
An MCP application that delivers curated positive and uplifting news stories using NewsAPI and Cohere LLM.
Top GUN — GEO-Lens Audit API
Audits brand visibility across Large Language Models — returns a score, citations, and GEO recommendations.
kabu-mcp
Japanese & Global stock market data for AI assistants. Real-time quotes, technical analysis (RSI/MACD/BB), market screening. The only MCP with Japanese stock support.
SwitchBot MCP Server
Control SwitchBot devices interactively using the SwitchBot API.
Transkribus MCP Server
MCP server for the Transkribus REST API — manage collections, documents, HTR/OCR recognition, models, and more. 290 tools across 22 resource domains.
Invoice Intelligence
A paid MCP server for invoice processing. Three tools: extract structured data from any invoice (pasted text, URL, or PDF), validate the extracted invoice for completeness and mathematical accuracy, and classify the expense into an accounting category with a suggested nominal account code and tax deductibility flag. Priced from $0.005 to $0.01 per call, paid in USDC on Base via x402. No API key required.
1Stay Hotel Booking
Transaction-complete hotel booking over MCP — 300K+ properties, real hotel confirmation numbers, loyalty points, secure checkout. Hotels are merchant of record. Builders set their own booking fee via Stripe Connect. Built on proven distribution infrastructure.
Google My Business MCP
Connect Google My Business to Claude or ChatGPT via Two Minute Reports MCP. Analyze profile views, customer actions, and top-performing locations to improve visibility and conversions.