Connect your agent

Exepad exposes its entire app lifecycle — create, edit, deploy, publish, observe — as an MCP server. Any MCP-capable agent (Claude Desktop, Claude Code, Cursor, or your own) can build and ship real, live web apps on your account.

Server: https://mcp.exepad.com/sse (SSE transport) · Auth: Authorization: Bearer exepad_dk_… · 29 tools (full list)

1. Get an API key from the Developer Console (Pro plan). For agent use, *:* scopes are convenient; pin the key to specific apps if you want to sandbox the agent — see Authentication.

2. Register the server with your client:

   Claude Desktop / Cursor

   Add to your mcp.json (Claude Desktop: Settings → Developer → Edit Config; Cursor: Settings → MCP):

   {
     "mcpServers": {
       "exepad": {
         "url": "https://mcp.exepad.com/sse",
         "headers": {
           "Authorization": "Bearer exepad_dk_..."
         }
       }
     }
   }

   Claude Code

   claude mcp add exepad --transport sse https://mcp.exepad.com/sse \
     --header "Authorization: Bearer exepad_dk_..."

   Custom agent (Python)

   import asyncio
   from mcp import ClientSession
   from mcp.client.sse import sse_client
   
   HEADERS = {"Authorization": "Bearer exepad_dk_..."}
   
   async def main():
       async with sse_client("https://mcp.exepad.com/sse", headers=HEADERS) as (r, w):
           async with ClientSession(r, w) as session:
               await session.initialize()
               tools = await session.list_tools()          # 29 exepad_* tools
               result = await session.call_tool(
                   "exepad_create_app",
                   {"name": "Recipe Box",
                    "description": "A recipe manager: title, ingredients, steps; browse as cards."},
               )
               print(result.content[0].text)
               # → {"app_id": "...", "edit_id": "...", "status": "processing", ...}
   
   asyncio.run(main())

3. Ask your agent to build something. For example, in Claude:

   Build me a habit tracker with daily check-ins and a streak counter, then publish it.

   A typical agent run chains: exepad_create_app → exepad_get_edit_status (poll until completed, ~5–9 min for a full-stack app) → exepad_publish → returns the live https://your-app.exepad.app URL.

What the agent can (and can’t) do

• Can: everything in the app lifecycle — create/clone/edit apps, upload knowledge-base files, deploy previews, publish/unpublish/rollback, set subdomains, read logs/health/errors, and call your apps’ backends (gateway tools).
• Can’t: manage API keys or webhooks (deliberately REST-only), or exceed the key’s scopes/app-pins — the server enforces them on every tool call.
• Destructive tools (exepad_delete_app, exepad_unpublish, exepad_remove_file, exepad_rollback) require an explicit confirm: true argument, so an agent can’t wipe an app on a whim.

Costs & limits

Tool calls are metered at $1 per million calls; AI generations (exepad_create_app, exepad_modify_app) are billed from your USD wallet like any other generation (pricing). Rate limits apply per key — creation is capped at 10 apps/hour (rate limits).

A2A instead of MCP?

If your agent framework speaks A2A rather than MCP, Exepad is also an A2A agent with public discovery at https://a2a.exepad.com/.well-known/agent-card.json. See the A2A guide.