SmitheryListed

CAIT

# CAIT - Core AI Toolkit A modular [Model Context Protocol (MCP)](https://modelcontextprotocol.io) server that extends AI assistants with practical capabilities: file I/O, a persistent Python REPL, AST-aware code analysis, semantic text search, document conversion, Wikipedia & arXiv tools, a persistent vector memory database, and other general utilities. A total of **37 tools across 9 modules**. Each module can be disabled independently via the `CAIT_DISABLE` environment variable. Made by AI for AI. ## Requirements - Python 3.11+ - Core: `fastmcp`, `chromadb` - Online research: `wikipedia-api`, `arxiv` - Document conversion: `docling` or `markitdown` - Scientific computing (optional, for REPL use): `sympy`, `scipy`, `matplotlib`, `plotly`, `vispy` ## Installation Install the requirements into your Python environment then copy [copilot-instructions.md](https://github.com/JacobBruce/CAIT/instructions/copilot-instructions.md) or [CLAUDE.md](https://github.com/JacobBruce/CAIT/instructions/CLAUDE.md) into the correct location (more details below). The instructions include general guidance for how to behave, how to use CAIT tools, and how to use the Firecrawl search tools. The instructions may need to be adapted to suite different setups. If you are working in a Python environment you may want to make use of this agent prompt: [python-coder.agent.md](https://github.com/JacobBruce/CAIT/agents/python-coder.agent.md). There is also [research-assistant.agent.md](https://github.com/JacobBruce/CAIT/agents/research-assistant.agent.md) for deep research. There is also a skill file called [project-planning.md](https://github.com/JacobBruce/CAIT/skills/project-planning.md) which is helpful for planning the implementation details of a project. The agent produces a PLAN.md file and TASKS.md file. ### Environment Variables | Variable | Default | Description | |----------|---------|-------------| | `CAIT_FILES_PATH` | `~/.cait/files/` | Directory for downloaded files and document cache | | `CAIT_DISABLE` | _(empty)_ | Comma-separated module names to exclude at startup (e.g. `wiki,arxiv`) | ## Client Configuration ### VS Code (GitHub Copilot) Add to your workspace `.vscode/mcp.json` or user `settings.json`: ```json { "servers": { "bitfreak/cait": { "type": "stdio", "command": "/absolute/path/to/.venv/bin/python", "args": ["-m", "cait.server"], "cwd": "/absolute/path/to/CAIT" } } } ``` > For user `settings.json`, nest the above under `"mcp": { ... }`. Copy [copilot-instructions.md](https://github.com/JacobBruce/CAIT/instructions/copilot-instructions.md) into your project's `.github/` folder to give Copilot guidance on using CAIT tools effectively. ### Claude Desktop Edit `claude_desktop_config.json`: - **macOS**: `~/Library/Application Support/Claude/claude_desktop_config.json` - **Linux**: `~/.config/claude/claude_desktop_config.json` - **Windows**: `%APPDATA%\Claude\claude_desktop_config.json` ```json { "mcpServers": { "bitfreak/cait": { "command": "/absolute/path/to/.venv/bin/python", "args": ["-m", "cait.server"], "env": { "PYTHONPATH": "/absolute/path/to/CAIT" } } } } ``` ### Claude Code ```bash claude mcp add cait -e PYTHONPATH=/absolute/path/to/CAIT \ -- /absolute/path/to/.venv/bin/python -m cait.server ``` Copy [CLAUDE.md](https://github.com/JacobBruce/CAIT/instructions/CLAUDE.md) to your project root (or to `~/.claude/CLAUDE.md` for global use) to give Claude Code guidance on using CAIT tools effectively. ## Recommended MCP Servers ### Firecrawl [Firecrawl](https://firecrawl.dev) is a web scraping and search API that pairs naturally with CAIT, adding powerful web search, full-page scraping, and site crawling. A free API key is available at [firecrawl.dev](https://firecrawl.dev). ### Serena [Serena](https://github.com/oraios/serena) provides many tools for semantic code retrieval and editing. Both CAIT and Serena include a similar memory system so it is recommended to disable one of them. ## Tool Reference ### File System — `fs` | Tool | De

First seen 2 Oct 2026. Evidence as of 2 Oct 2026.