Documentation

Complete guide to the Intuition MCP monorepo

Overview

The Intuition MCP monorepo ships 2 MCP servers exposing 17 tools that let AI assistants like Claude query the Intuition knowledge graph, compute trust scores, and verify on-chain reputation data.

Intuition MCP

8 tools · View playground

Official Model Context Protocol server for the Intuition knowledge graph. Provides read access to accounts, relationships, atoms, and list data across the Intuition network.

Trust Score MCP

9 tools · View playground

MCP server for graph-based trust computation. Implements EigenTrust, AgentRank, sybil detection, and composite scoring over the Intuition attestation graph.

Installation

1. Clone the monorepo

git clone https://github.com/intuition-box/mcp.git
cd mcp

2. Install dependencies

Run npm install at the repository root. Workspaces are configured so every package gets its dependencies resolved in one pass.

npm install

3. Configure environment

# apps/playground/.env.local
NEXT_PUBLIC_INTUITION_GRAPH_URL=https://graph.intuition.systems/graphql

4. Start the playground

cd apps/playground
npm run dev

The playground runs at http://localhost:3000 and gives you an interactive UI to explore both MCP servers.

Monorepo Structure

mcp/
├── packages/
│   ├── mcp-general/       # Intuition MCP server (knowledge graph)
│   └── mcp-trust/         # Trust Score MCP server (EigenTrust, AgentRank)
├── apps/
│   └── playground/        # Next.js interactive playground & docs
├── package.json           # Workspace root
└── README.md

Claude Desktop Integration

Configure both MCP servers

Add the following to your Claude Desktop config file to register both servers.

Config file location

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "intuition": {
      "command": "node",
      "args": [
        "/absolute/path/to/mcp/packages/mcp-general/dist/index.js"
      ],
      "env": {
        "NEXT_PUBLIC_INTUITION_GRAPH_URL": "https://graph.intuition.systems/graphql"
      }
    },
    "trust-score": {
      "command": "node",
      "args": [
        "/absolute/path/to/mcp/packages/mcp-trust/dist/index.js"
      ],
      "env": {
        "NEXT_PUBLIC_INTUITION_GRAPH_URL": "https://graph.intuition.systems/graphql"
      }
    }
  }
}

Replace /absolute/path/to/mcp with the actual path where you cloned the repository. Restart Claude Desktop after saving changes.

Tools Reference

Intuition MCP8 tools

`get-account-info`

Retrieve detailed profile and metadata for an Intuition account.

`get-followers`

List accounts that follow a given account.

`get-following`

List accounts that a given account is following.

`get-inbound-relations`

Retrieve inbound relation edges pointing to an account.

`get-outgoing-edges`

Retrieve outgoing relation edges originating from an account.

`search-account-ids`

Search for account identifiers matching a query string.

`search-atoms`

Search atoms in the knowledge graph by label or content.

`search-lists`

Search curated lists within the Intuition network.

Trust Score MCP9 tools

`eigentrust`

Compute EigenTrust scores for a set of agents in the trust graph.

`agentrank`

Rank agents using the AgentRank algorithm over attestation edges.

`composite-score`

Generate a weighted composite trust score combining multiple trust signals.

`sybil-simulation`

Run a sybil-resistance simulation to detect coordinated fake identities.

`transitive-trust`

Calculate transitive trust between two agents across intermediate paths.

`network-stats`

Return aggregate statistics about the trust network topology.

`trust-path`

Find the strongest trust path between a source and target agent.

`get_sync_status`

Returns the current auto-sync cron job status -- whether it is running, next scheduled run, last run time, and whether the last run succeeded.

`run_sync`

Manually trigger a graph sync from Intuition GraphQL to Neo4j. Fetches latest attestations and updates the trust graph. Returns sync result with nodes created, edges created, and duration.

HTTP API Endpoints

When the playground dev server is running you can also call the tools via HTTP.

GET /api/trust-score

curl "http://localhost:3000/api/trust-score?address=0x..."

GET /api/attestations

curl "http://localhost:3000/api/attestations?subject=0x...&limit=50"

POST /api/mcp

curl -X POST http://localhost:3000/api/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "tool": "verifyCredential",
    "params": {
      "address": "0x...",
      "claim": "expert-in-defi"
    }
  }'

Algorithm Documentation

The Trust Score MCP server implements seven core algorithm areas for computing reputation, influence, and sybil resistance over the Intuition attestation graph. Each algorithm is documented in detail in the full technical specification.

EigenTrust

Iterative power-iteration algorithm computing global trust scores by propagating normalized attestation weights across the graph until convergence. Based on the Kamvar, Schlosser, Garcia-Molina (2003) paper for sybil-resistant reputation.

Read full specification →

AgentRank

PageRank variant with stake-weighted edges for influence ranking. Produces per-node influence scores alongside network-level metrics including Gini coefficient and Shannon entropy for inequality and diversity measurement.

Read full specification →

Composite Scoring

Weighted combination of EigenTrust (0.4), AgentRank (0.3), and Transitive Trust (0.3) into a normalized 0–100 score with confidence indicators. Supports per-query weight overrides and batch computation.

Read full specification →

Multi-Hop Transitive Trust

Personalized trust propagation through multi-hop paths with configurable per-hop decay, stake weighting, and predicate-specific multipliers. Supports both targeted pairwise queries and full outgoing network scans.

Read full specification →

Sybil Resistance

Simulation-based detection that injects synthetic sybil clusters into the graph, measures their impact on trust scores, then cleans up. Produces a resistance score quantifying how well the network withstands coordinated manipulation.

Read full specification →

Graph Indexer

Pipeline syncing Intuition attestation data from the GraphQL API into Neo4j using cursor-based pagination and batch MERGE operations. Tracks sync health metrics including duration, error counts, and graph size.

Read full specification →

Predicate Weights

Configurable weight system where each attestation predicate (trusts, follows, is-qualified) carries a numeric multiplier scaling its contribution to trust calculations. Supports per-query overrides and custom predicate registration.

Read full specification →

Resources

Intuition Systems

Learn about the Intuition protocol

intuition.systems →

Model Context Protocol

Official MCP specification

modelcontextprotocol.io →

GitHub Repository

View source code & contribute

github.com/intuition-box/mcp →

MCP Directory

Explore & try all available servers

Browse MCP Directory →