Home

MCP Explorer

Welcome to the MCP Explorer developer documentation!

MCP Explorer is a state-of-the-art, local-first agentic query interface built with Python and LangChain. It uses a ReAct (Reasoning and Acting) loop powered by local large language models (such as gemma4:e4b via Ollama) to autonomously invoke tools, query external APIs, search databases, and format results.

The system is equipped with Model Context Protocol (MCP) clients to connect to remote or local MCP servers dynamically and query resources in real time.

---

🌟 Key Features

• Local-First Agent Core: Run LLMs locally using Ollama (defaulting to the gemma4:e4b model) with zero latency or remote API dependency.
• Dynamic Model Context Protocol (MCP) Loader: Automatically discover, load, and authenticate tools from multiple MCP servers dynamically using configurations in mcpConfig.json.
• Standard Tools Extension: Load and dynamically configure parameters (like custom user agents or endpoints) for classic tools (Wikipedia, Arxiv) defined dynamically in toolsConfig.json.
• Interactive Chat UI: A responsive, real-time streaming web dashboard built using Chainlit for developer conversations.
• Observability Pipeline: Automatic tracking of every message, prompt, LLM invocation, tool execution, token usage, and error boundary using Langfuse.
• Dockerized Stack: Run the entire system—including database backends, workers, and visualization dashboards—using a simple ./docker-run.sh script.

---

🗺️ High-Level System Architecture

The workflow follows a standard ReAct loop:

    [User Inputs Query]
             │
             ▼
   [Chainlit App Session]
             │
             ▼
      [ReAct Agent] <───(Reads Configs: env / toolsConfig.json)
             │
     ┌───────┴───────┐
     ▼               ▼
[Run Standard]  [Run MCP Server]
   [Tools]          [Tools]
     │               │
     └───────┬───────┘
             ▼
     [Agent Formulates] ───(Callback)───► [Langfuse Traces Logged]
       [Response]
             │
             ▼
     [Stream to UI]