Tru8 Evidence Research MCP server

Tru8 — AI-Powered Evidence Research Platform

Tru8 organises evidence so you can decide what matters. Submit a claim or URL, and Tru8 retrieves evidence from the open web and specialist APIs, classifies it by tier (primary / reporting / commentary) and type (data / official / news / analysis / opinion / academic), decomposes claims into elements, maps evidence to those elements, and presents the full landscape — structured, not summarised.

Mission: "We organise; you decide."

Quick Start

Prerequisites

• Docker & Docker Compose
• Python 3.11+
• Node.js 20+

1. Start Infrastructure

docker-compose up -d    # PostgreSQL 16, Redis 7, Qdrant

2. Backend

cd backend
python -m venv venv
source venv/bin/activate    # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env        # Edit with your API keys
alembic upgrade head
uvicorn app.main:app --reload    # http://localhost:8000

3. Frontend

cd web
npm install
npm run dev    # http://localhost:3000

Project Structure

tru8/
├── backend/              # FastAPI backend + evidence pipeline
│   ├── app/
│   │   ├── api/v1/       # REST endpoints (checks, agent, verify, payments)
│   │   ├── core/         # Config, auth, pricing, manifest signing
│   │   ├── models/       # SQLModel schemas (Check, Claim, Evidence, ClaimConsensus)
│   │   ├── pipeline/     # Two-phase evidence pipeline
│   │   ├── services/     # API adapters, search, payments, consensus
│   │   └── utils/        # Classifiers, deduplication, query planning
│   ├── alembic/          # Database migrations
│   ├── tru8_mcp/         # MCP server package for AI agents
│   └── tests/            # 990 unit + integration tests
├── web/                  # Next.js 14 frontend
│   ├── app/              # Pages (dashboard, developers, public reports)
│   ├── components/       # Evidence views, claim map, marketing
│   └── lib/              # API client, tiers, diagnostics
├── shared/               # Shared TypeScript types
├── docker-compose.yml    # PostgreSQL, Redis, Qdrant, MinIO (optional)
└── .claude/CLAUDE.md     # Full engineering context

Pipeline

Two-phase pipeline with user claim selection gate (article mode):

Phase 1:  Ingest → Extract claims → Classify article → Rank claims
          [PAUSE — user selects claims]

Phase 2:  Fact-check lookup → Decompose into elements → Retrieve evidence →
          Score relevance → Classify tier/type → Map evidence to elements →
          Derive orientation → Complete

Post:     Coverage recovery (Stage 5.1), video recommendations, auto-archiving

Evidence flows through web search plus specialist source adapters:

Six Evidence Views

API & Agent Access

Tru8 exposes a developer API and MCP server for AI agents.

Agent Commerce Gateway

# Quick check (~15s, heuristic classification)
curl -X POST https://api.trueight.com/api/v1/agent/quick \
  -H "X-API-Key: $TRU8_API_KEY" \
  -d '{"claim": "UK inflation is 3.2%"}'

Shortened. The full README is on GitHub.