<!-- mcp-name: io.github.fgasparetto/chipsai-mcp -->

ChipsAI MCP Server

MCP (Model Context Protocol) server for ChipsBot — manage chatbots, conversations, documents, bot-to-bot routing, RAG configuration, and AI models from Claude Code, Claude Desktop, or any MCP client.

Requirements

• Python 3.11+
• uv (recommended) or pip
• A ChipsBot account (sign up)

Quick Start

No installation needed with uv:

uv run --script server.py

Or install manually:

pip install "mcp[cli]" httpx
python server.py

Configuration

The server uses environment variables for authentication. API key is the recommended method — generate one from your ChipsBot dashboard.

<details> <summary>Legacy: username/password authentication</summary>

If you don't have an API key, you can use username/password instead:

</details>

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "chipsai": {
      "command": "uvx",
      "args": ["chipsai-mcp"],
      "env": {
        "CHIPSAI_API_KEY": "chipsai_your_api_key_here"
      }
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "chipsai": {
      "command": "uvx",
      "args": ["chipsai-mcp"],
      "env": {
        "CHIPSAI_API_KEY": "chipsai_your_api_key_here"
      }
    }
  }
}

Available Tools

Chatbot Management

Documents (RAG)

Conversations

Widget History

Chat

Bot-to-Bot Connections

RAG Configuration

User & Models

RAG Pipeline

ChipsBot supports a full Retrieval-Augmented Generation pipeline configurable per-bot:

• Semantic routing (L1): pgvector + Jina Embeddings v3 — routes queries to the best specialist based on cosine similarity (HNSW index)
• HyDE: for sparse/short queries, generates a hypothetical answer with Haiku and re-embeds it for better retrieval
• Chunk injection (L2): at response time, injects only the top-K relevant KB chunks instead of the full prompt — reduces token usage, improves quality
• Reranking: optional Jina cross-encoder reranker (jina-reranker-v2-base-multilingual) applied after cosine retrieval
• Chunking strategies: char (fixed size), paragraph (semantic \n\n split), sentence (.!? split)
• Document upload: PDF/DOC/DOCX parsed via LlamaParse, extracted text stored as KB

Use get_rag_config / update_rag_config to tune all parameters per-bot.

Bot-to-Bot Routing

An orchestrator bot can route questions to specialist bots based on role/description. The orchestrator detects [ROUTE:uuid] tags in its own response and delegates to the matching specialist, passing recent chat history as context.

Use connect_bot to link specialists to an orchestrator, list_bot_connections to inspect the routing table, and update_bot_connection to adjust roles or toggle connections on/off.

Credit System

ChipsAI uses a credit-based pricing model:

Credit packages: 150 credits for €5 | 700 for €20 | 2000 for €50. Credits never expire. Bring your own API key to use any model for free (no credits consumed).

Usage Examples

Once configured, use natural language in Claude:

• "List my chatbots"
• "Create a chatbot called Support Bot"
• "Upload the product catalog PDF to my chatbot"
• "Send a test message to my chatbot"
• "Show analytics for the last 7 days"
• "Change the chatbot model to Claude Sonnet 4.6"
• "What's my credit balance?"
• "What AI models are available?"
• "Connect the billing bot as a specialist of my main orchestrator"
• "List all specialist bots connected to my orchestrator"
• "Show the RAG config for my chatbot"
• "Set the RAG threshold to 0.5 and enable reranking"
• "Enable L2 chunk injection with top_k=5"

Authentication

API Key (recommended): Set CHIPSAI_API_KEY with a key generated from your dashboard. The key is sent as a Bearer token — no token management needed.

JWT (legacy): If using username/password, tokens are obtained via JWT and refreshed transparently.

License

MIT