How It Works

Conjra bridges the gap between natural language instructions and cloud infrastructure APIs using the Model Context Protocol (MCP).

The MCP architecture

MCP is an open protocol that lets AI applications expose "tools" — functions the AI can call on behalf of the user. Conjra implements an MCP server that registers infrastructure tools with your AI editor.

You say: "use conjra to create a supabase project" │ ▼ AI Editor (Claude Code, Windsurf, etc.) Recognizes "conjra" → calls create_supabase_project │ ▼ JSON-RPC call over stdio Conjra MCP Server - Receives tool call - Validates input against Zod schema - Loads credentials from local vault - Calls provider API - Returns structured result │ ▼ HTTPS request Provider API (api.supabase.com) - Creates the resource - Returns IDs, URLs, keys │ ▼ Your AI receives the result

Tool discovery

When your editor starts, it reads Conjra tool definitions. Each tool has a name, description, input schema (Zod), and execute function. The AI decides which tool to call based on your natural language request.

Credential management

  1. You run conjra add <provider> in your terminal
  2. Conjra prompts for API keys
  3. Keys are encrypted with AES-256-GCM and stored in ~/.conjra/vault/
  4. When a tool needs credentials, Conjra decrypts them in memory, makes the API call, and discards them
  5. Your keys are machine-locked — encrypted with a key derived from your machine hostname, platform, and architecture

Next steps

Edit this page on GitHub
Previous← Quick StartNextAI Editors→