vCon MCP Server

A Model Context Protocol (MCP) server for storing, managing, and analyzing IETF vCon (Virtual Conversation) data with AI assistants.

Overview

This MCP server provides a standardized way for AI assistants like Claude to interact with conversation data using the IETF vCon (Virtual Conversation) format. It combines the Model Context Protocol's tool-based interaction model with Supabase's powerful PostgreSQL backend to create a fully spec-compliant conversation data management system.

What is vCon?

vCon (Virtual Conversation) is an IETF standard for representing conversations in a portable, interoperable format. Think of it as "PDF for conversations" - a standardized container for:

• Conversations from any medium (voice, video, text, email)

• Participants with identity and privacy controls

• AI Analysis from transcription, sentiment, summarization, etc.

• Attachments like documents, images, or related files

• Privacy markers for consent and redaction

What is MCP?

The Model Context Protocol (MCP) enables AI assistants to use external tools and data sources. This server implements MCP to give AI assistants the ability to create, search, analyze, and manage conversation data.

Key Features

• ✅ IETF vCon Compliant - Implements draft-ietf-vcon-vcon-core-02 specification (v0.4.0)

• ✅ MCP Integration - 35 tools for AI assistants to manage conversation data

• ✅ Redesigned Contract Tools - Additive vcon_fetch, vcon_capabilities, vcon_search, vcon_taxonomy, and describe_response_shape tools for predictable envelopes, explicit payload control, and discovery-first clients

• ✅ REST API - Full HTTP REST API with CRUD, discovery, filtered reads, search, tags, and analytics surfaces

• ✅ Database Analytics - Comprehensive analytics for size, growth, content patterns, and health monitoring

• ✅ Large Database Support - Smart response limiting, metadata-only options, and memory-safe queries

• ✅ OpenTelemetry Observability - Full traces, metrics, and structured logs with console or OTLP export

• ✅ Query Prompts - 9 pre-built prompts to guide effective searching and retrieval:

  • Exact tag matching (e.g., "find angry customers from June")

  • Semantic search (e.g., "find frustrated users")

  • Keyword search (e.g., "find conversations mentioning refund")

  • Multi-criteria queries with step-by-step guidance

• ✅ Supabase Backend - Powerful PostgreSQL database with REST API

• ✅ Redis Caching - Optional high-performance cache layer for 20-50x faster reads

• ✅ Type-Safe - Full TypeScript implementation with Zod validation

• ✅ Plugin Architecture - Extensible plugin system for custom functionality

• ✅ Privacy-Ready - Plugin hooks for implementing consent, redaction, and compliance

• ✅ Advanced Search - Unified vcon_search plus specialized legacy search tools:

  • Basic filtering (subject, parties, dates)

  • Full-text keyword search (dialog, analysis, parties)

  • Semantic search (AI embeddings for meaning-based search)

  • Hybrid search (combines keyword and semantic)

• ✅ Tag Filtering - Filter search results by tags via attachments of type tags

• ✅ Content Indexing - Searches dialog bodies and analysis content (encoding='none')

• ✅ Real-time-ready - Supabase realtime subscriptions are available at the database layer (not yet exposed via MCP tools)

• ✅ Conserver Integration - Compatible with vCon conserver for chain processing

Quick Start

Prerequisites

• Node.js 18+

• npm or yarn

• Supabase account (sign up free)

Installation

# Clone the repository
git clone https://github.com/vcon-dev/vcon-mcp.git
cd vcon-mcp

# Install dependencies
npm install

# Set up environment variables
cp .env.example .env
# Edit .env with your Supabase credentials

# Build the project
npm run build

# Run tests
npm test

# Start the server
npm run dev

Configure with Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "vcon": {
      "command": "node",
      "args": ["/path/to/vcon-mcp/dist/index.js"],
      "env": {
        "SUPABASE_URL": "your-project-url",
        "SUPABASE_SERVICE_ROLE_KEY": "your-service-role-key",
        "SUPABASE_ANON_KEY": "your-anon-key",
        "OPENAI_API_KEY": "your-openai-key"
      }
    }
  }
}

Note: REDIS_URL is optional. If provided, enables high-performance caching for 20-50x faster reads. See Redis-Supabase Integration Guide.

Database backend: Defaults to Supabase (DB_TYPE=supabase). MongoDB is also supported via DB_TYPE=mongodb — see docs/mongodb/setup.md.

Restart Claude Desktop and start using vCon tools!

Docker Deployment

The vCon MCP Server is available as a Docker image for easy deployment.

Quick Start with Docker

# Pull the image
docker pull public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main

# Run the server (using service role key for full access)
docker run -p 3000:3000 \
  -e SUPABASE_URL=https://your-project.supabase.co \
  -e SUPABASE_SERVICE_ROLE_KEY=your-service-role-key \
  -e SUPABASE_ANON_KEY=your-anon-key \
  -e MCP_HTTP_STATELESS=true \
  public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main

Note: You need either SUPABASE_SERVICE_ROLE_KEY (recommended for full access) or SUPABASE_ANON_KEY (for restricted access). The service role key bypasses Row Level Security.

Image Tags

Running Scripts

The Docker image includes all utility scripts:

# Show available commands
docker run --rm public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main help

# Check database status
docker run --rm \
  -e SUPABASE_URL=your-url \
  -e SUPABASE_SERVICE_ROLE_KEY=your-service-role-key \
  -e SUPABASE_ANON_KEY=your-anon-key \
  public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main script check-db-status

# Run embeddings with OpenAI
docker run --rm \
  -e SUPABASE_URL=your-url \
  -e SUPABASE_SERVICE_ROLE_KEY=your-service-role-key \
  -e SUPABASE_ANON_KEY=your-anon-key \
  -e OPENAI_API_KEY=your-openai-key \
  public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main script embed-vcons --provider=openai

# Run embeddings with Azure OpenAI
docker run --rm \
  -e SUPABASE_URL=your-url \
  -e SUPABASE_SERVICE_ROLE_KEY=your-service-role-key \
  -e SUPABASE_ANON_KEY=your-anon-key \
  -e AZURE_OPENAI_EMBEDDING_ENDPOINT=https://your-resource.openai.azure.com \
  -e AZURE_OPENAI_EMBEDDING_API_KEY=your-azure-key \
  public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main script embed-vcons --provider=azure

Docker Compose

version: '3.8'
services:
  vcon-mcp:
    image: public.ecr.aws/r4g1k2s3/vcon-dev/vcon-mcp:main
    ports:
      - "3000:3000"
    environment:
      - SUPABASE_URL=${SUPABASE_URL}
      - SUPABASE_SERVICE_ROLE_KEY=${SUPABASE_SERVICE_ROLE_KEY}
      - SUPABASE_ANON_KEY=${SUPABASE_ANON_KEY}
      - MCP_HTTP_STATELESS=true

Multi-Client Support

Enable stateless mode for multiple simultaneous client connections:

-e MCP_HTTP_STATELESS=true

Connecting to Local Services

When connecting to services on your host machine (like local Supabase), use host.docker.internal:

-e SUPABASE_URL=http://host.docker.internal:54321

See Docker Deployment Guide for complete documentation.

Common Operations

The project includes npm scripts for common database and data management operations:

Database Status & Analysis

# Comprehensive database status (recommended)
npm run db:status

# Quick vCon count check
npm run db:check

# Daily count analysis to identify gaps
npm run db:analyze

Data Loading

# Sync vCons from S3/legacy sources
npm run sync:vcons

# Full sync (vCons + embeddings + tags)
npm run sync

# Continuous sync (watches for new data)
npm run sync:continuous

Backup & Restore

# Backup database
npm run db:backup

# Restore from backup
npm run db:restore

Embeddings Management

# Continuously generate embeddings for vCons missing them
npm run sync:embeddings

# Check embedding coverage
npm run embeddings:check

Testing

# Test database tools
npm run test:db

# Test search functionality
npm run test:search

# Test tag system
npm run test:tags

For more details on scripts and advanced options, see scripts/README.md.

Transport Options

The vCon MCP Server supports multiple transport mechanisms for connecting AI assistants:

stdio Transport (Default)

Standard input/output transport for CLI-based AI assistants like Claude Desktop.

# Default mode - no configuration needed
npm run dev

Or explicitly in .env:

MCP_TRANSPORT=stdio

HTTP Transport with Streamable HTTP

HTTP server mode enables browser-based clients and remote connections using the MCP Streamable HTTP specification (2025-03-26).

Features:

• ✅ Stateful or stateless session management

• ✅ SSE streaming for real-time responses

• ✅ JSON-only mode for simple request/response

• ✅ CORS support for browser clients

• ✅ DNS rebinding protection for security

• ✅ Session resumability (optional)

Configuration:

# .env
MCP_TRANSPORT=http
MCP_HTTP_HOST=127.0.0.1
MCP_HTTP_PORT=3000

# Optional: Stateless mode (no session tracking)
# MCP_HTTP_STATELESS=true

# Optional: JSON-only responses (no SSE streaming)
# MCP_HTTP_JSON_ONLY=true

# Optional: CORS for browser clients
# MCP_HTTP_CORS=true
# MCP_HTTP_CORS_ORIGIN=*

# Optional: DNS rebinding protection
# MCP_HTTP_DNS_PROTECTION=true
# MCP_HTTP_ALLOWED_HOSTS=localhost,127.0.0.1
# MCP_HTTP_ALLOWED_ORIGINS=http://localhost:3000

Starting the HTTP Server:

npm run dev

The server will start on http://127.0.0.1:3000 (default) and log:

✅ vCon MCP Server running on HTTP
🌐 Listening on: http://127.0.0.1:3000
📡 Mode: Stateful

Client Connection:

# Step 1: Initialize MCP connection
curl -i -X POST http://127.0.0.1:3000 \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"my-client","version":"1.0.0"}}}'

# Extract the Mcp-Session-Id from response headers

# Step 2: Send MCP requests with session ID
curl -X POST http://127.0.0.1:3000 \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: <session-id-from-step-1>" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'

# GET request to establish SSE stream (for notifications)
curl -X GET http://127.0.0.1:3000 \
  -H "Mcp-Session-Id: <your-session-id>"

# DELETE request to close session
curl -X DELETE http://127.0.0.1:3000 \
  -H "Mcp-Session-Id: <your-session-id>"

Session Management:

• Stateful mode (default): Server generates and tracks session IDs

• Stateless mode (MCP_HTTP_STATELESS=true): No session tracking, each request is independent

See MCP Streamable HTTP Specification for protocol details.

REST API

When running in HTTP transport mode, the server exposes a full RESTful HTTP API with parity to all MCP tools. Both interfaces share the same VConService business logic, plugin hooks, and database queries.

Path layout: the MCP HTTP endpoint is mounted at / (root), and the REST API is mounted at REST_API_BASE_PATH (default /api/v1). They share the same port — that's why the curl examples above hit / for MCP but /api/v1/... for REST.

Endpoints (30+)

Configuration

# REST API settings (optional - defaults work for most cases)
REST_API_BASE_PATH=/api/v1     # Base path for endpoints
REST_API_ENABLED=true          # Enable/disable REST API
CORS_ORIGIN=*                  # CORS allowed origins

# API Key Authentication
API_KEYS=key1,key2             # Comma-separated valid API keys
API_AUTH_REQUIRED=true         # Set to false to disable auth

Quick Example

# Health check
curl http://localhost:3000/api/v1/health

# Create a vCon (requires API key)
curl -X POST http://localhost:3000/api/v1/vcons \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-api-key" \
  -d '{"vcon":"0.4.0","subject":"Support Call","parties":[{"name":"Agent","tel":"+1111"}]}'

# Search by keyword
curl "http://localhost:3000/api/v1/vcons/search/content?q=billing+issue" \
  -H "Authorization: Bearer your-api-key"

# Discover attachment purposes before reading by category
curl "http://localhost:3000/api/v1/discovery/attachments/purposes" \
  -H "Authorization: Bearer your-api-key"

# Read only dealer-info attachments for one vCon
curl "http://localhost:3000/api/v1/vcons/123e4567-e89b-12d3-a456-426614174000/attachments?purpose=dealer_info" \
  -H "Authorization: Bearer your-api-key"

When a classification lives in attachments or analysis, discover the live values first and read through those filtered surfaces. For attachments, prefer purpose because it is the spec field; use attachment type only as a legacy compatibility fallback. Use tags when the classification genuinely lives in the tags attachment.

See the complete REST API Reference for detailed documentation.

Available MCP Tools

The server exposes 35 MCP tools, grouped by purpose. See docs/api/tools.md for full schemas.

Recommended For New Clients

The preferred starting point for new clients is the additive contract-oriented tool family:

For new LLM-built or hand-written clients, start with vcon_capabilities, then vcon_taxonomy, then build on vcon_search and vcon_fetch.

Are The Older Tools Still Useful?

Yes. The older tools are still useful for:

• Backward compatibility with existing clients already built around get_vcon, search_vcons, and the older search variants

• Low-level or specialized flows where a client intentionally wants the older response shapes or narrower operations

• Incremental migration, where working code can adopt the redesigned tools gradually instead of switching in one cutover

The legacy tools remain supported, but they are no longer the recommended starting point for new client work because their response envelopes and payload shapes are less predictable.

CRUD

Search

Contract & Discovery

Tags

Analytics

Schema & Examples

Database Inspection

Tool Categories & Configuration

Tools are organized into categories that can be enabled or disabled for different deployment scenarios. By default, all categories are enabled.

Categories

Configuration Options

Configure via environment variables:

# Option 1: Use a preset profile
MCP_TOOLS_PROFILE=readonly   # Options: full, readonly, user, admin, minimal

# Option 2: Enable specific categories only
MCP_ENABLED_CATEGORIES=read,write,schema

# Option 3: Disable specific categories (starts with all enabled)
MCP_DISABLED_CATEGORIES=analytics,infra

# Option 4: Disable individual tools
MCP_DISABLED_TOOLS=delete_vcon,analyze_query

Deployment Profiles

Example: Read-Only Deployment

{
  "mcpServers": {
    "vcon": {
      "command": "node",
      "args": ["/path/to/vcon-mcp/dist/index.js"],
      "env": {
        "SUPABASE_URL": "your-project-url",
        "SUPABASE_ANON_KEY": "your-anon-key",
        "MCP_TOOLS_PROFILE": "readonly"
      }
    }
  }
}

Available MCP Resources

The server exposes URI-based resources for direct reads:

• vcon://v1/discovery/attachments/purposes – discovered attachment purposes with counts

• vcon://v1/discovery/attachments/types – discovered legacy attachment types with counts

• vcon://v1/discovery/analysis/types – discovered analysis types with counts

• vcon://v1/vcons/{uuid} – full vCon JSON

• vcon://v1/vcons/{uuid}/metadata – metadata only

• vcon://v1/vcons/{uuid}/parties – parties array

• vcon://v1/vcons/{uuid}/dialog – dialog array

• vcon://v1/vcons/{uuid}/attachments – attachments array

• vcon://v1/vcons/{uuid}/attachments/purpose/{purpose} – attachments filtered by purpose

• vcon://v1/vcons/{uuid}/attachments/type/{type} – attachments filtered by legacy type

• vcon://v1/vcons/{uuid}/analysis – analysis array

• vcon://v1/vcons/{uuid}/analysis/type/{type} – analysis filtered by type

• vcon://v1/vcons/{uuid}/transcript – transcript analysis (filtered)

• vcon://v1/vcons/{uuid}/summary – summary analysis (filtered)

• vcon://v1/vcons/{uuid}/tags – tags as object (parsed)

For attachment- or analysis-backed classifications, start with the discovery resources and then read the matching filtered surface. For attachments, prefer purpose; use type only for compatibility with older datasets. Prefer tags only when the data is actually stored as tags.

Use Cases

Contact Centers

• Capture and analyze customer calls

• Generate automatic transcripts

• Track agent performance and sentiment

• Maintain compliance audit trails

Sales Teams

• Record sales conversations

• Extract action items and follow-ups

• Analyze conversation patterns

• Generate meeting summaries

Research

• Collect conversation datasets

• Analyze communication patterns

• Generate insights from dialogue

• Build ML training data

Compliance & Legal

• Maintain conversation archives

• Apply redaction for privacy

• Track consent and permissions

• Generate audit reports

Architecture

Basic Architecture

┌─────────────────────┐
│   AI Assistant      │  (Claude, ChatGPT, etc.)
│   (Client)          │
└──────────┬──────────┘
           │ MCP Protocol (stdio)
           │
┌──────────▼──────────┐
│   vCon MCP Server   │  (This project)
│                     │
│  ┌───────────────┐  │
│  │ MCP Tools     │  │  - create_vcon
│  │               │  │  - add_analysis
│  └───────┬───────┘  │  - search_vcons
│          │          │  - etc.
│  ┌───────▼───────┐  │
│  │ vCon Queries  │  │  - CRUD operations
│  │               │  │  - Validation
│  └───────┬───────┘  │  - Type checking
│          │          │
└──────────┼──────────┘
           │ Supabase Client
           │
┌──────────▼──────────┐
│    Supabase         │
│  (PostgreSQL)       │
│                     │
│  ┌───────────────┐  │
│  │ vCon Tables   │  │  - vcons
│  │               │  │  - parties
│  └───────────────┘  │  - dialog
│  ┌───────────────┐  │  - analysis
│  │ pgvector      │  │  - attachments
│  │ (embeddings)  │  │
│  └───────────────┘  │
└─────────────────────┘

With Redis Caching (Optional)

For high-performance deployments, add Redis as a cache layer:

┌─────────────────────────────────────────────────┐
│                  AI Assistant                    │
└─────────────────────┬───────────────────────────┘
                      │ MCP Protocol
┌─────────────────────▼───────────────────────────┐
│              vCon MCP Server                     │
│                                                   │
│  ┌───────────────────────────────────────────┐  │
│  │         Cache-First Reads                  │  │
│  │  Redis (hot) → Supabase (cold fallback)   │  │
│  └───────────────────────────────────────────┘  │
└─────────────┬───────────────────┬───────────────┘
              │                   │
      ┌───────▼───────┐   ┌───────▼───────┐
      │  Redis Cache  │   │   Supabase    │
      │  (Optional)   │   │  (Permanent)  │
      │               │   │               │
      │ - Fast reads  │   │ - Source of   │
      │ - TTL expiry  │   │   truth       │
      │ - Auto cache  │   │ - Full CRUD   │
      └───────────────┘   └───────────────┘

Enable caching by setting REDIS_URL environment variable. See Redis-Supabase Integration Guide for details.

Performance: Redis caching provides 20-50x faster reads for frequently accessed vCons.

Observability

Built-in OpenTelemetry instrumentation provides production-ready monitoring:

Features

• Distributed Tracing: Full request lifecycle tracing with spans for every operation

• Business Metrics: Track vCon operations, search patterns, and usage

• Performance Metrics: Monitor query duration, cache hit rates, and latency

• Structured Logging: JSON logs with automatic trace context correlation

• Flexible Export: Console (development) or OTLP (production) exporters

Quick Start

# Development (console export)
OTEL_ENABLED=true
OTEL_EXPORTER_TYPE=console

# Production (OTLP collector)
OTEL_ENABLED=true
OTEL_EXPORTER_TYPE=otlp
OTEL_ENDPOINT=http://localhost:4318

Key Metrics

• vcon.created.count - vCons created

• vcon.deleted.count - vCons deleted

• vcon.search.count - Searches performed (by type)

• tool.execution.duration - Tool execution time

• db.query.count - Database queries

• cache.hit / cache.miss - Cache performance

Collector Examples

Testing Setup (Recommended):

# Start Jaeger backend using docker-compose
./jaeger/start-jaeger.sh

# Configure .env
OTEL_ENABLED=true
OTEL_EXPORTER_TYPE=otlp
OTEL_ENDPOINT=http://localhost:4318

# View traces at http://localhost:16686

Manual Docker Setup:

# Jaeger (all-in-one)
docker run -d -p 4318:4318 -p 16686:16686 jaegertracing/all-in-one:latest

# View traces at http://localhost:16686

See Observability Guide for complete documentation.

Project Structure

For the authoritative source layout (including src/api/, src/server/, src/services/, src/prompts/, src/hooks/ and the route/handler split), see the File Structure section in CLAUDE.md. That document is kept in sync with the code; this README intentionally does not duplicate the tree.

Documentation

For Users

• Getting Started - Quick start guide for using the server

• Docker Deployment - Complete Docker deployment guide

• Query Prompts Guide - How to use search and retrieval prompts

• Search Tools Guide - Search strategies and tools

• Tag Management Guide - Tagging and organization

• RLS Multi-Tenant Guide - Row Level Security setup for multi-tenant isolation

For Developers

• Building Guide - Step-by-step implementation from scratch

• Search Optimization - Vector search performance

• Plugin Development - Creating custom plugins

For LLMs and AI Systems

• Database Documentation Index - Start here - Overview of all LLM documentation

• Database Architecture for LLMs - Complete database design and structure

• Database Quick Start for LLMs - Practical code examples and patterns

• Database Schema Visual Reference - Visual entity relationship diagrams

API Reference

• REST API - Full HTTP REST API with parity to all MCP tools

• Tools API - MCP tools reference

• Prompts API - MCP prompts reference

• Resources API - MCP resources reference

• Types - TypeScript type definitions

Technical Reference

• Quick Reference - Critical spec corrections checklist

• Implementation Corrections - Detailed spec compliance guide

• Corrected Schema - Database schema reference

• Migration Guide - Migrating existing code

IETF Specifications

• IETF vCon Core Spec (-00, in repo) - v0.3.0 baseline shipped with this repo

• IETF vCon Core Spec (current draft) - Latest -02 (v0.4.0) on the IETF datatracker

• vCon Consent Draft - Privacy and consent

• vCon Lifecycle Draft - Lifecycle management

• vCon Quick Start - vCon basics

• vCon Adapter Guide - Building adapters

Development

Running Locally

# Start Supabase (if using local)
supabase start

# Run in development mode
npm run dev

# Run tests
npm test

# Run spec compliance tests
npm run test:compliance

# Launch MCP test console (interactive)
npm run test:console

# Build for production
npm run build

# Lint code
npm run lint

Testing

The project includes comprehensive tests:

• Unit tests - Type validation, query functions

• Integration tests - End-to-end vCon operations

• Compliance tests - IETF spec conformance

# Run all tests
npm test

# Run with coverage
npm test -- --coverage

# Run specific test file
npm test tests/vcon-compliance.test.ts

Database Setup

The project uses Supabase with a carefully designed schema:

• 8 tables for vCon data model

• 25 indexes for query performance

• Row Level Security for multi-tenancy (configurable tenant extraction from attachments)

• pgvector for semantic search

• Realtime subscriptions enabled

See the Building Guide for complete database setup instructions. See RLS Multi-Tenant Guide for enabling multi-tenant isolation.

IETF vCon Specification Compliance

This implementation targets draft-ietf-vcon-vcon-core-02 (vCon v0.4.0), including:

Core Objects

• ✅ vCon container with all required fields

• ✅ Party objects with complete metadata

• ✅ Dialog objects (recording, text, transfer, incomplete)

• ✅ Analysis objects with vendor and schema fields

• ✅ Attachment objects with proper references

• ✅ Group objects for multi-party conversations

Data Types

• ✅ Correct field names (e.g., schema not schema_version)

• ✅ Required vs optional fields properly enforced

• ✅ String-based body fields (not object types)

• ✅ No default encoding values

• ✅ Proper type constraints

Privacy & Security

• ✅ Redaction support

• ✅ Consent tracking

• ✅ Party privacy markers

• ✅ Secure attachment handling

API Examples

Creating a vCon

const vcon = await createVCon({
  subject: "Customer Support Call",
  parties: [
    {
      name: "Alice Agent",
      mailto: "alice@support.example.com",
      role: "agent"
    },
    {
      name: "Bob Customer",
      tel: "+1-555-0100",
      role: "customer"
    }
  ]
});

Adding Analysis

await addAnalysis(vconUuid, {
  type: "transcript",
  vendor: "OpenAI",
  product: "Whisper-1",
  schema: "v1.0",
  body: "Full transcript text...",
  encoding: "none",
  dialog: [0]  // References first dialog
});

Searching vCons

const results = await searchVCons({
  subject: "billing",
  partyName: "Alice",
  startDate: "2025-01-01",
  endDate: "2025-01-31"
});

Roadmap

Phase 1: Core Implementation ✅

• IETF vCon type definitions

• Supabase database schema

• Basic CRUD operations

• MCP server implementation

• Validation and testing

Phase 2: Advanced Features 🚧

• Semantic search with pgvector

• Real-time subscriptions

• Batch operations

• Export/import formats

Phase 3: Enterprise Features 📋

• Multi-tenant support (RLS with configurable tenant extraction)

• Advanced privacy controls

• Audit logging

• Performance optimization

Phase 4: Integrations 📋

• Twilio adapter

• Zoom adapter

• Slack adapter

• Microsoft Teams adapter

Extending the Server

The vCon MCP Server is highly extensible, supporting multiple ways to add custom functionality:

Extension Options

Quick Start: Add a Custom Resource

// src/resources/index.ts
export function getCoreResources(): ResourceDescriptor[] {
  return [
    // ... existing resources ...
    {
      uri: 'vcon://v1/analytics/summary',
      name: 'Analytics Summary',
      description: 'Overall conversation analytics',
      mimeType: 'application/json'
    }
  ];
}

Quick Start: Add a Custom Tool

// src/tools/analytics-tools.ts
export const analyticsTool = {
  name: 'get_analytics',
  description: 'Get conversation analytics',
  inputSchema: {
    type: 'object' as const,
    properties: {
      period: { type: 'string', enum: ['7d', '30d', '90d'] }
    },
    required: ['period']
  }
};

export async function handleGetAnalytics(input: any): Promise<ToolResponse> {
  // Implementation
}

Quick Start: Create a Plugin

import { VConPlugin, RequestContext } from 'vcon-mcp/hooks';

export default class MyPlugin implements VConPlugin {
  name = 'my-plugin';
  version = '1.0.0';
  
  // Register custom tools
  registerTools(): Tool[] {
    return [/* your tools */];
  }
  
  // Register custom resources
  registerResources(): Resource[] {
    return [/* your resources */];
  }
  
  // Lifecycle hook example
  async afterCreate(vcon: VCon, context: RequestContext): Promise<void> {
    console.log(`Created vCon: ${vcon.uuid}`);
  }
}

Loading Plugins

# Set environment variable
VCON_PLUGINS_PATH=@mycompany/vcon-plugin,./local-plugin.js
VCON_LICENSE_KEY=your-license-key-if-required

# Run server with plugins
npm run dev

Complete Extension Documentation

• Extension Guide - Comprehensive guide with examples

• Extension Quick Reference - Fast lookup and decision guide

• Plugin Development - Complete plugin documentation

• Custom Tools - Tool development guide

Contributing

We welcome contributions! Here's how to get started:

1. Fork the repository

2. Create a feature branch (git checkout -b feature/amazing-feature)

3. Make your changes

4. Run tests (npm test)

5. Commit your changes (git commit -m 'Add amazing feature')

6. Push to the branch (git push origin feature/amazing-feature)

7. Open a Pull Request

Contribution Guidelines

• Follow the existing code style

• Add tests for new features

• Update documentation

• Ensure IETF spec compliance

• Reference spec sections in comments

License

MIT License - see LICENSE file for details

Resources

Project Links

• GitHub: vcon-mcp

• Issues: Bug reports & feature requests

• Discussions: Community discussions

External Links

• IETF vCon Working Group: datatracker.ietf.org/wg/vcon

• Model Context Protocol: modelcontextprotocol.io

• Supabase: supabase.com

• vCon GitHub: github.com/vcon-dev

Support

• 🐛 Bug Reports: GitHub Issues

• 💬 Discussions: GitHub Discussions

• 📖 Documentation: mcp.conserver.io

Acknowledgments

• IETF vCon Working Group for the specification

• Anthropic for the Model Context Protocol

• Supabase for the amazing PostgreSQL platform

• Contributors who helped build and improve this project

Built with ❤️ for the conversation intelligence community

Making conversations accessible, analyzable, and actionable with AI