Claude Desktop
Anthropic's desktop app. MCP-native — edit one JSON file and you're done.
Connect an AI agent to your vault
Pick your assistant. Each walkthrough is 5 steps: create a BoxOwl token, configure the agent, test a query.
Supported assistants
Claude Code (CLI)
The Claude developer CLI. One-shot install via npx @boxowl/skills add.
Cursor
The AI-first code editor. MCP-native; @boxowl/skills wires it up.
VS Code Copilot
GitHub Copilot's MCP support in VS Code (uses servers not mcpServers).
Hermes Agent (Nous Research)
Nous's Docker-distributed Python agent with native MCP support. Drop a boxowl: block into ~/.hermes/config.yaml.
ChatGPT
Custom GPT with BoxOwl's REST API as an Action. No MCP support yet, but the agent token works as a bearer.
Other MCP hosts
The @boxowl/skills CLI also
configures Windsurf, Codex (OpenAI), and Gemini CLI in one shot. Run
npx @boxowl/skills add --platform=<id> with one of:
windsurf, codex, gemini-cli. Per-host setup walkthroughs for those land as separate pages
when the post-launch per-vendor variant work lands.
Background
- Agent tokens (REST API reference) — the token format, scopes vocabulary, and the
Authorization: Bearerusage every walkthrough below builds on. - API quick start — register, authenticate, read a vault category in three curl calls.
- MCP transport docs are coming with the post-launch
boxowl-mcpPython package; until then, the REST agent API is the canonical interface (see the per-host walkthrough or the full API reference).