MCP Client: Agentic Chatbot with Tool-Use and Streamlit UI

Overview

This project is an agentic chatbot client that can use external tools (such as SQLite and Puppeteer servers) to answer user queries, automate workflows, and perform multi-step reasoning. It features both a command-line interface and a modern Streamlit web UI. The chatbot is powered by an LLM (via Groq API) and can chain tool calls, passing outputs from one tool as inputs to the next.

---

Features

• Agentic LLM Chatbot: Uses an LLM to interpret user queries and decide when/how to use tools.
• Tool Chaining: Supports multi-step workflows where the output of one tool is used as input for another.
• SQLite Tooling: Query, update, and manage a SQLite database via an MCP server.
• Puppeteer Tooling: Automate browser actions (navigate, click, fill, screenshot, evaluate JS, etc.) via Puppeteer MCP server.
• Streamlit Web UI: Chat with the agent, view logs, and see tool outputs in a modern web interface.
• Command-Line Mode: Interact with the agent via the terminal.
• Logging: All actions and errors are logged to app.log for transparency and debugging.

---

Folder Structure

app_streamlit.py      # Streamlit web app for chat and tool use
main.py               # Main agentic chatbot logic (CLI entrypoint)
requirements.txt      # Python dependencies
servers_config.json   # Configuration for MCP tool servers
app.log               # Log file (auto-generated)
test.db               # SQLite database (auto-generated/used by sqlite server)
.env                  # Environment variables (API keys, etc.)

---

Setup Instructions

1. Python Version

• Windows users: You must use Python 3.10 or 3.11. Python 3.13+ is not supported for async subprocesses (required for tool servers).
• Download Python 3.11

2. Clone and Install

# Clone the repo (if not already)
cd path\to\your\projects
# Create and activate a virtual environment
py -3.11 -m venv venv
.\venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt

3. Environment Variables

• Create a .env file in the project root with your LLM API key:

  LLM_API_KEY=your_groq_api_key_here

4. Configure Tool Servers

• Edit servers_config.json to define which MCP servers to use. Example:

  {
    "mcpServers": {
      "sqlite": {
        "command": "uvx",
        "args": ["mcp-server-sqlite", "--db-path", "./test.db"]
      },
      "puppeteer": {
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
      }
    }
  }

• Ensure you have Node.js and the required MCP servers installed globally or accessible via npx/uvx.

---

Usage

Command-Line Chatbot

python main.py

• Type your message and press Enter.
• Type exit or quit to stop.
• The agent will use tools as needed and log all actions to app.log.

Streamlit Web App

streamlit run app_streamlit.py

• Open the provided local URL in your browser.
• Chat with the agent, view logs, and see tool outputs.
• If you are on Windows with Python 3.13+, you will see an error message (use Python 3.11).

---

Tool Chaining (Agentic Mode)

• The agent can execute multiple tools in sequence, passing results between them.
• Example LLM tool call (handled automatically):

  [
    {"tool": "puppeteer_navigate", "arguments": {"url": "https://news.google.com/"}, "result_var": "page_content"},
    {"tool": "summarize_text", "arguments": {"text": "$page_content"}, "result_var": "summary"},
    {"tool": "write_query", "arguments": {"query": "INSERT INTO summaries (text) VALUES ('$summary')"}}
  ]

• The agent will resolve $page_content and $summary automatically.

---

Logs

• All actions, tool calls, and errors are logged to app.log.
• In the Streamlit app, you can view logs in the "Show Log Output" expander.

---

Troubleshooting

• Python 3.13+ on Windows: Not supported for async subprocesses. Use Python 3.11.
• Tool server fails to start: Check your servers_config.json and ensure uvx, npx, and the MCP servers are installed and on your PATH.
• No tools available: Make sure servers are running and initialized (see logs for errors).
• API errors: Check your .env and API key.

---

Credits

• Model Context Protocol (MCP)
• Streamlit
• Groq API

---

License

This project is for educational and research purposes. See individual tool/server repos for their licenses.