VectorForge MCP Server

A Model Context Protocol (MCP) server for VectorForge APIs, providing cryptographic integrity, audit trails, and confidence scoring for AI applications.

Installation

npm install -g @vectorforge-ai/mcp-server

Or use directly with npx:

npx @vectorforge-ai/mcp-server

Quick Start

1. Get Your API Key

Sign up at vectorforge.ai and create an API key.

2. Configure Claude Desktop

Add to your Claude Desktop configuration:

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

{
  "mcpServers": {
    "vectorforge": {
      "command": "npx",
      "args": ["@vectorforge-ai/mcp-server"],
      "env": {
        "VF_API_KEY": "vf_prod_your_key_here",
        "VF_API_BASE_URL": "https://api.vectorforge.ai"
      }
    }
  }
}

3. Restart Claude Desktop

Claude will now have access to all VectorForge tools.

Available Tools

| Tool | Description | |------|-------------| | vf.register | Register content and get a cryptographic DIVT | | vf.verify | Verify content against a registered DIVT | | vf.get_bundle | Get a full verification bundle for a DIVT | | vf.score.privacy | Privacy-preserving confidence scoring (no raw content sent) | | vf.score.full | Full semantic confidence scoring with Groq judge | | vf.stream_events | Stream audit events via SSE | | vf.get_worldstate | Get a single worldstate record | | vf.list_worldstate | List worldstate records with filters | | vf.prompt_receipt.create | Create a receipt for AI interactions | | vf.rag_snapshot.create | Freeze your knowledge base state | | vf.agent_action.log | Log agent/automation actions | | vf.keys.create | Create new API keys | | vf.keys.revoke | Revoke an API key |

Examples

Register Content

Tool: vf.register
Arguments:
  object_id: "doc-123"
  data_type: "document_v1"
  hash_mode: "content"
  content: "This is my important document."

Returns a DIVT with cryptographic signatures (ECDSA P-521 + ML-DSA-65).

Verify Content

Tool: vf.verify
Arguments:
  divt_id: "019abc12-3456-7890-abcd-ef0123456789"
  content: "This is my important document."

Returns verification status: hash validity, signature validity, revocation status.

Create Prompt Receipt

Tool: vf.prompt_receipt.create
Arguments:
  prompt: "What is the capital of France?"
  response: "The capital of France is Paris."
  model: "claude-3"
  register_divt: true

Score Answer Confidence

Tool: vf.score.full
Arguments:
  query: "What are the side effects of aspirin?"
  answer: "Common side effects include stomach upset and heartburn."
  evidence:
    - text: "Aspirin may cause stomach irritation, heartburn, and nausea."
      similarity: 0.92

Log Agent Action

Tool: vf.agent_action.log
Arguments:
  action: "approve_refund"
  actor: "support-agent-1"
  params:
    order_id: "ORD-123"
    amount: 49.99
  register_divt: true

Environment Variables

| Variable | Description | Required | |----------|-------------|----------| | VF_API_KEY | Your VectorForge API key | Yes | | VF_API_BASE_URL | API URL (e.g., https://api.vectorforge.ai) | Yes |

Integration Examples

LangChain

import { MCPTool } from '@langchain/community/tools/mcp';

const vfTool = new MCPTool({
  serverCommand: 'npx',
  serverArgs: ['@vectorforge-ai/mcp-server'],
  toolName: 'vf.register',
  env: {
    VF_API_KEY: process.env.VF_API_KEY,
    VF_API_BASE_URL: process.env.VF_API_BASE_URL,
  }
});

Direct MCP Client

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { StdioClientTransport } from '@modelcontextprotocol/sdk/client/stdio.js';

const transport = new StdioClientTransport({
  command: 'npx',
  args: ['@vectorforge-ai/mcp-server'],
  env: {
    VF_API_KEY: 'vf_prod_...',
    VF_API_BASE_URL: 'https://api.vectorforge.ai',
  }
});

const client = new Client(
  { name: 'my-app', version: '1.0.0' },
  { capabilities: {} }
);

await client.connect(transport);

const tools = await client.listTools();
const result = await client.callTool('vf.register', {
  object_id: 'test-123',
  data_type: 'test_v1',
  hash_mode: 'content',
  content: 'Test content',
});

Use Cases

AI Audit Trail

Record every AI interaction with cryptographic proof:

User prompt -> vf.prompt_receipt.create -> Immutable record with DIVT

RAG Provenance

Track which knowledge base version produced each answer:

Index build -> vf.rag_snapshot.create -> Sealed corpus version
Query -> vf.score.full -> Confidence with evidence links

Agent Accountability

Log every automated action for compliance:

Agent decision -> vf.agent_action.log -> Tamper-evident audit trail

Architecture

+-------------------+     +------------------------+     +-------------------+
|  Claude Desktop   |---->|  VectorForge MCP       |---->|  VectorForge      |
|  (or LangChain)   |     |  Server (local)        |     |  API (remote)     |
+-------------------+     +------------------------+     +-------------------+
       User                npm package (free)              Your API calls

The MCP server runs locally on your machine and makes API calls to VectorForge's cloud service. Content is sent to the API for server-side canonicalization, hashing, and signing.

Development

git clone https://github.com/VectorForgeAI/API.git
cd API/mcp
npm install
npm run build
node dist/index.js

Related

• Node.js SDK
• Python SDK
• CLI

License

MIT