AI Assistant with MCP, LangChain, and Groq

A small terminal chat app that connects a Groq LLM to multiple MCP (Model Context Protocol) servers through mcp-use and LangChain. You type natural language; the agent picks tools (browser automation, search, etc.) and runs them for you.

Features

• Multi-server MCP — Configured via browser_mcp.json (Playwright browser, Airbnb, DuckDuckGo search).
• Groq — Fast inference via langchain-groq (ChatGroq).
• Conversation memory — MCPAgent keeps context across turns until you clear it.
• Robust tool calls — Custom middleware strips null optional arguments so strict MCP servers (e.g. Playwright) do not reject JSON null for optional string fields.
• Browser search hints — System instructions steer the model toward reliable patterns (e.g. Google: navigate to https://www.google.com/search?q=... instead of typing into ambiguous homepage refs).

Requirements

Requirement	Notes
Python	3.13+ (see pyproject.toml)
uv	Used to create the venv and run the app
Node.js	npx must work; MCP servers in this project start via npx
Groq API key	Groq Console

Quick start

1. Clone the repository (use the exact URL from your GitHub repo page).

2. Install dependencies

   cd mcpdemos
   uv sync
   

3. Create a .env file in the project root (this file is gitignored):

   GROQ_API_KEY=your_groq_api_key_here
   

4. Run the chat loop

   uv run app.py
   

   First run may download MCP packages through npx; allow network access.

Usage

• Type your request at the You: prompt (e.g. open a site, search, or combine steps).
• exit, quit, or bye — Leave the app.
• clear, clear history, or reset — Clear the agent’s conversation memory.

Configuration

browser_mcp.json

Defines MCP servers in the standard mcpServers shape. This project includes:

Server	Purpose
playwright	@playwright/mcp — browser navigate, click, type, snapshot, etc.
airbnb	@openbnb/mcp-server-airbnb
duckduckgo-search	duckduckgo-mcp-server

You can add/remove servers or change args; restart the app after edits.

app.py (knobs you might change)

• config_file — Path to the MCP JSON config (default: browser_mcp.json).
• ChatGroq — model, max_tokens, temperature.
• MCPAgent — max_steps, memory_enabled, additional_instructions (browser/search behavior).

Project layout

Path	Role
app.py	Async chat loop, MCP client, middleware, agent setup
browser_mcp.json	MCP server definitions
main.py	Minimal entry stub (primary demo is app.py)
pyproject.toml / uv.lock	Dependencies and lockfile

Troubleshooting

• Repository not found (git) — Confirm the remote URL matches your GitHub repo slug exactly (including any trailing . in the repo name if you created it that way).
• 429 from Groq — Rate limit; wait and retry or check Groq rate limits.
• Playwright / browser errors — Ensure Node and npx work; first run may install @playwright/mcp. For Google searches, the app is tuned to prefer direct search?q= URLs to avoid filling non-input elements.
• Telemetry — mcp-use may log anonymized telemetry; set MCP_USE_ANONYMIZED_TELEMETRY=false in the environment if you want it off.

License

Specify your license here (e.g. MIT) or add a LICENSE file.