Mistr. Agent

What is Mistr. Agent?

Mistr. Agent is an autonomous MCP (Model Context Protocol) client for Mistral AI models. It runs as a web application on localhost, connects Mistral language models with real-world tools via MCP servers, and is intended for users who want AI-driven multi-step task completion across local and web environments.

How to use Mistr. Agent?

Clone the repository, install dependencies with npm install, copy .env.example to .env.local and add your API keys, then run npm run dev. Access the interface at http://localhost:3000. Configure MCP servers by editing mcp-config.json—pre‑configured entries for Perplexity search and Desktop Commander are provided.

Key features of Mistr. Agent

• Autonomous multi‑turn, multi‑tool task execution with context maintenance
• Self‑correction of tool usage when parameters are incorrect
• Human‑in‑the‑loop approval for sensitive operations
• Dynamic loading and discovery of MCP servers and their capabilities
• Terminal command execution and local filesystem operations
• Web search via Perplexity AI integration
• Dark mode and conversation history interface

Use cases of Mistr. Agent

• Research Assistant: Search the web, compile findings into local files, and generate summaries
• Code Helper: Search local repositories, analyze code, execute tests, and explain results
• System Administrator: Monitor processes, execute commands, and manage files with natural language
• Content Creator: Research topics online and create structured documents from the findings

FAQ from Mistr. Agent

What models does Mistr. Agent support?

All Mistral chat models are supported; the application has been tested with Mistral 8B and Mistral Large. You can configure your preferred model in the settings.

Can I add custom MCP servers?

Yes. Install or set up the server process, add a new entry to the servers object in mcp-config.json, and restart the development server. The client will automatically discover the new tools.

How does the security and approval system work?

The client provides a UI toggle for "Auto Approve". In manual mode every tool call requires user approval; in auto mode calls execute without confirmation. Additional security features include rate limiting, detection of dangerous operations, and detailed audit logging.

What are the system requirements?

Node.js v16.x or higher, 4 GB RAM minimum (8 GB recommended), 500 MB storage, and an internet connection for API access. Smaller local models like Mistral 8B can be used if available.

What license does Mistr. Agent use?

The project is licensed under the MIT License.