Connect Your IDE

The first thing to do with Qosm is connect its MCP server to your coding environment. This gives your LLM tools to typecheck, save, and run Qosm code — so it can write agents for you.

Sign in and get a token

1. Go to qosm.io/workspace/invite and sign in (or create an account with your invite code)
2. Go to Settings
3. Under Personal Access Tokens, enter a name (e.g., "Claude Code") and click Generate
4. Copy the token immediately — it's shown only once

Add the MCP server

The server URL is:

https://mcp.qosm.io/mcp

Claude Code

claude mcp add qosm \
  --transport http \
  https://mcp.qosm.io/mcp \
  -- \
  --header "Authorization: Bearer YOUR_TOKEN"

Cursor

1. Open Settings → MCP
2. Click Add new MCP server
3. URL: https://mcp.qosm.io/mcp
4. Header: Authorization: Bearer YOUR_TOKEN

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "qosm": {
      "url": "https://mcp.qosm.io/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

What your LLM can now do

Once connected, the LLM has these tools:

| Tool | What it does | |------|-------------| | get_projects | List your projects | | get_project_context | See a project's code and capabilities | | get_project_functions | List functions with types and effects | | qosm_check | Typecheck code — catches errors before saving | | qosm_push | Save code to a project | | qosm_call_function | Run a specific function | | qosm_init | Create a new project |

The important one is qosm_check. Your LLM will use it in a loop — write code, check it, fix errors, check again — until the compiler is happy. This is what makes the workflow reliable: the LLM iterates against a real typechecker, not just its own intuition.

Next, let's configure capabilities — the permissions your agent code will have.