Overview

This MCP (Model Context Protocol) server provides a lightweight bridge between Quickchat AI agents and your knowledge base. Deploying the server gives agents live, on-demand access to documents, product data, internal policies, or any other structured/unstructured content you host. Instead of relying solely on pre-baked prompts or cached embeddings, Quickchat can request context during a conversation and receive up-to-date information from your systems.

The server implements the MCP-compatible HTTP interface expected by Quickchat agents and translates incoming context requests into searches or lookups against your configured data sources (e.g., vector stores, databases, file systems, or search APIs). This enables real-time conversational apps that remain synchronized with your latest content and reduces hallucination by returning verifiable, source-linked context snippets.

Features

Implements Model Context Protocol to serve live context to Quickchat AI agents
Connects to multiple knowledge sources: vector DBs, search APIs, SQL/NoSQL, filesystem
Pluggable retrieval logic (semantic search, keyword search, hybrid)
Authentication/secret validation for secure agent-server communication
Health and metrics endpoints for monitoring
Example connectors and deployment-ready Docker image

Installation / Configuration

Clone the repository and run with Docker or locally via Node.js (or the supplied runtime). Replace environment variables with values for your environment.

Clone and run with Docker

git clone https://github.com/incentivai/quickchat-ai-mcp.git
cd quickchat-ai-mcp

# Build image
docker build -t quickchat-mcp .

# Run container (example)
docker run -d \
  -p 3000:3000 \
  -e MCP_SECRET="your-mcp-secret" \
  -e KB_PROVIDER="vector" \
  -e VECTOR_ENDPOINT="https://your-vector-db.example" \
  --name quickchat-mcp \
  quickchat-mcp

Run locally (Node.js)

cd quickchat-ai-mcp
npm install
# copy .env.example to .env and edit
npm start

Key environment variables

Variable	Description	Example
MCP_SECRET	Shared secret used to validate incoming Quickchat requests	MCP_SECRET=“s3cr3t”
PORT	HTTP port for the MCP server	PORT=3000
KB_PROVIDER	Type of knowledge source (vector, search, sql, fs)	KB_PROVIDER=“vector”
VECTOR_ENDPOINT	Endpoint for vector DB or search API (if applicable)	VECTOR_ENDPOINT=“https://vector.example/api”
LOG_LEVEL	Logging verbosity (info, debug, warn)	LOG_LEVEL=“info”

Configuration files and connector modules live in the repo under a config/ and connectors/ directory. Update or extend connectors to adapt to your data sources.

Available Resources

The server exposes a small set of HTTP endpoints used by Quickchat agents and for operations:

GET /health — simple healthcheck returning 200 OK
POST /mcp/context — main MCP endpoint: accepts a model context request and returns context blocks
GET /metrics — optional Prometheus-style metrics (if enabled)
POST /admin/reload — reloads connector configs without restarting (protected)

Example request to /mcp/context

curl -X POST https://mcp.example.com/mcp/context \
  -H "Authorization: Bearer your-mcp-secret" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "quickchat-large",
    "query": "How do I reset my password?",
    "conversation": [{"role":"user","text":"I forgot my password"}],
    "max_results": 4
  }'

Example response (trimmed)

{
  "context": [
    {
      "id": "doc:123",
      "title": "Password Reset Guide",
      "text": "To reset your password, go to /settings/security and click Reset Password...",
      "score": 0.92,
      "source": "kb://internal-articles/123"
    }
  ],
  "meta": {
    "provider": "vector",
    "took_ms": 45
  }
}

Use Cases

Support chatbots with live product information: return current stock, pricing, or specs from your product database during customer conversations.
Internal knowledge assistants: let employees query up-to-date policies, onboarding docs, or runbook snippets without shipping secrets into the model.
Real-time troubleshooting: integrate logs or monitoring links so agents can cite recent incidents or metrics in responses.
E-commerce and catalog search: augment agent prompts with context-rich product descriptions and user-specific recommendations fetched at runtime.

Concrete example: customer support

User asks a question about a recent order.
Quickchat agent sends an MCP context request containing the user message and order id.
MCP server queries your order database and knowledge base, returning the order status and relevant FAQ snippets.
The agent composes a response that references the order status and links to the exact policy doc.

Getting Started Tips

Start with a single connector (e.g., a vector DB) and verify the /mcp/context flow using curl.
Use a short-lived development secret for testing, then rotate to a secure secret/store in production.
Add source and provenance metadata in context blocks so agents can cite where information came from.
Monitor /metrics and logs to understand query latencies and tune retrieval parameters (top-k, rerankers, filters).

Repository and further docs: https://github.com/incentivai/quickchat-ai-mcp

Quickchat AI MCP Server: Live Knowledge Base Access