@odei/mcp-server

MCP server that gives Claude Desktop access to ODEI's constitutional world model -- a knowledge graph with 7-layer safety guardrails.

Quick Start | Tools | Architecture | API Reference | Contributing

Auto-updated 3x/day · Personal Agent App · DAOrg · 2026-03-11 16:00 UTC

What is this?

ODEI is a constitutional AI operating system built on a Neo4j knowledge graph. This MCP server exposes ODEI's world model and guardrail system to any MCP-compatible AI client -- Claude Desktop, Cursor, Windsurf, or your own agent.

What you get:

• World Model Query -- search a production knowledge graph with 59 entity types across 6 domains

• Guardrail Check -- validate any agent action against constitutional safety constraints before execution

• Trust Signals -- rapid confidence scoring for claims, entities, and trends

• Smart Contract Audit -- EVM address analysis cross-referenced against the knowledge graph

The MCP server runs locally and connects to the ODEI API. The /api/v2/ endpoints (guardrail, world model query) require an API key — see Authentication. The public world model projection at /api/worldmodel/live is free to use.

Quick Start

npm registry: @odei/mcp-server is not yet published to npm. Install directly from GitHub (Option 2 or 3 below).

Option 1: npx from GitHub

npx github:odei-ai/mcp-odei

Option 2: Clone and run locally

git clone https://github.com/odei-ai/mcp-odei.git
cd mcp-odei
npm install && npm run build
node dist/index.js

Option 3: Add to Claude Desktop

Add to your ~/.claude/mcp.json (or claude_desktop_config.json):

{
  "mcpServers": {
    "odei": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-odei/dist/index.js"],
      "env": {
        "ODEI_API_KEY": "your-key-here"
      }
    }
  }
}

Then restart Claude Desktop. You'll see 4 new tools available.

Try it — ask Claude: "What are ODEI's current strategic priorities?"

Claude responds using odei_world_model_query:

Found 8 active strategy-layer nodes:
→ Ship ODEI Personal AI OS (Q1 2026, 42% complete)
→ Reach 2,000 $ODAI holders (847 current)
→ x402 payment gateway — LIVE on Base
→ Agent marketplace expansion — 5/8 platforms active

Constitutional layers queried: FOUNDATION → VISION → STRATEGY
Response time: 145ms | Cache: HIT

Option 4: Smithery

npx @smithery/cli install @odei/mcp-server

Authentication

The MCP server uses the ODEI API. Most tools route through /api/worldmodel/live (public, no auth required). The /api/v2/ endpoints require authentication:

API Key — include in your MCP config env:

{ "ODEI_API_KEY": "your-key-here" }

The key is sent as X-Api-Key: <key> on every request. Contact odei@odei.ai for access.

Without a key, the MCP server still works: odei_world_model_query, odei_guardrail_check, and odei_world_model_signal all run against the public /api/worldmodel/live endpoint locally — no v2 API call is made.

Why ODEI Memory?

Tools

odei_world_model_query

Search the knowledge graph by keyword, node type, or domain layer.

Query: "health goals"
Domain: VISION
Type: goal
Limit: 10

Example response:

{
  "ok": true,
  "totalNodes": 91,
  "matchedCount": 3,
  "nodes": [
    {
      "id": "n_a1b2c3d4",
      "title": "Health",
      "type": "Goal",
      "domain": "VISION",
      "labels": ["Entity", "OdeiNode", "Goal"]
    }
  ],
  "connectedEdges": [
    {
      "source": "n_a1b2c3d4",
      "target": "n_e5f6g7h8",
      "type": "PURSUES_GOAL"
    }
  ]
}

Parameters:

odei_guardrail_check

Validate an action against constitutional guardrails. Returns APPROVED, DENIED, or NEEDS_REVIEW.

Action: "Transfer 5 ETH to external wallet"
Severity: high
Domain: EXECUTION

Example response:

{
  "ok": true,
  "verdict": "NEEDS_REVIEW",
  "action": "Transfer 5 ETH to external wallet",
  "warnings": [
    {
      "source": "Financial Guardrail",
      "note": "Financial action detected — verify amount and recipient"
    }
  ],
  "guardrailsChecked": 3,
  "principlesChecked": 5,
  "valuesChecked": 10
}

Parameters:

odei_world_model_signal

Rapid trust and relevance scoring. Faster than a full query when you just need a confidence number.

Claim: "ODAI token has 2000 holders"
Category: fact_check

Example response:

{
  "ok": true,
  "signal": "MODERATE",
  "confidence": 0.45,
  "evidenceNodes": 3,
  "topMatches": [
    {
      "title": "$ODAI Token",
      "type": "Asset",
      "domain": "FOUNDATION",
      "relevanceScore": 0.67
    }
  ]
}

Parameters:

odei_smart_contract_audit

Analyze an EVM contract address. Cross-references known ODEI contracts and the knowledge graph.

Address: "0x0086cff0c1e5d17b19f5bcd4c8840a5b4251d959"
Chain: base

Example response:

{
  "ok": true,
  "address": "0x0086cff0c1e5d17b19f5bcd4c8840a5b4251d959",
  "riskLevel": "SAFE",
  "riskScore": 0,
  "knownContract": {
    "name": "$ODAI Token (Flaunch V1_2)",
    "chain": "base",
    "verified": true
  },
  "findings": [
    {
      "severity": "info",
      "title": "Known ODEI Contract",
      "detail": "This is $ODAI Token (Flaunch V1_2) on base. Verified: true."
    }
  ],
  "explorerUrl": "https://basescan.org/address/0x0086cff0c1e5d17b19f5bcd4c8840a5b4251d959"
}

Parameters:

Architecture

ODEI's world model is a constitutional knowledge graph stored in Neo4j. The graph is organized into 6 hierarchical domains:

    FOUNDATION          Core identity, values, guardrails, capabilities
         |
       VISION           Goals, north stars, business objectives
         |
      STRATEGY          OKRs, strategies, key results
         |
      TACTICS           Projects, milestones, risks, roadmaps
         |
     EXECUTION          Tasks, processes, systems, time blocks
         |
       TRACK            Observations, metrics, decisions, signals

Graph Stats

Node Types (selected)

Foundation: Human, AI, Partnership, Value, Principle, Guardrail, Capability, Constraint, Policy, Resource

Vision: Goal, North Star, Vision, Business, Season

Strategy: Strategy, Objective, Key Result, Initiative

Tactics: Project, Milestone, Risk, Program, Roadmap, Opportunity

Execution: Task, Area, System, Process, Time Block

Track: Observation, Decision, Signal, Pattern, Metric, Insight, Fact, Event, Action

Safety Layers

The guardrail system provides 7 layers of protection:

1. Constitutional Principles -- immutable values that cannot be violated

2. Financial Guards -- spending limits, transfer verification

3. Temporal Guards -- time-of-day restrictions (late-night caution)

4. Health Guards -- workload and wellness constraints

5. Identity Guards -- impersonation and credential protection

6. Relationship Guards -- social and reputational safety

7. Destructive Operation Guards -- prevent bulk deletions and bypasses

Configuration

Environment Variables

Transport Modes

stdio (default): For local clients like Claude Desktop and Cursor.

node dist/index.js

SSE: For remote or web-based clients.

node dist/index.js --sse
# Listening on http://localhost:4100/sse

API Reference

The MCP server communicates with ODEI's API:

Base URL: https://api.odei.ai

Public endpoints support CORS and return JSON. Rate limit: 20 requests/minute per IP.

Authentication for

X-Api-Key: your-key-here

All v2 endpoints return 401 UNAUTHORIZED without a valid key. The MCP tools automatically use the public /api/worldmodel/live projection when no API key is configured, so they still work unauthenticated.

MCP Resources

The server also exposes MCP resources for passive context:

On-Chain Contracts

ODEI has verified deployments on Base:

The odei_smart_contract_audit tool knows about these and will flag them as verified.

Development

git clone https://github.com/odei-ai/mcp-odei.git
cd mcp-odei
npm install
npm run dev          # Run with tsx (auto-reload)
npm run build        # Compile TypeScript
npm run typecheck    # Type checking only
npm run lint         # ESLint

Testing locally with Claude Desktop

1. Build: npm run build

2. Add to ~/.claude/mcp.json:

{
  "mcpServers": {
    "odei-dev": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-odei/dist/index.js"]
    }
  }
}

3. Restart Claude Desktop.

Contributing

See CONTRIBUTING.md for development setup and guidelines.

Links

• Website: odei.ai

• API & Dashboard: api.odei.ai

• Token: GeckoTerminal

• Twitter: @odei_ai

• GitHub: github.com/odei-ai

License

MIT -- ODEI Symbiosis