Skip to main content

Execution and Prompts MCP Servers

Aisle includes two built-in MCP servers accessible from the Connectors page. They let external MCP clients - Claude Code, Cursor, or any compatible tool - call your Aisle prompts and workflows as tools.

The two servers​

Prompts Server - exposes a prompt as an MCP native prompt. Your client receives the rendered template with variables filled in. No model call happens inside Aisle; your client sends the rendered message to whichever model it uses.

Execution Server - exposes a prompt or workflow as an MCP tool. Calling it runs the full execution inside Aisle - model call, workflow logic, everything - and returns the result to your client.

Use the Prompts Server when your client handles model selection and you want to use Aisle-maintained templates. Use the Execution Server when you want to call a specific prompt or workflow and get back its output.

Connecting​

Both servers are listed on the Connectors page under the built-in server section.

  1. Navigate to Connectors
  2. Find Execution Server or Prompts Server
  3. Click to open the connection setup
  4. Select which prompt or workflow to expose
  5. Generate a bearer token - this scopes the connection to that specific resource
  6. Copy the connection details into your MCP client

Each connection is scoped to one prompt or workflow. To expose multiple resources, create multiple connections with separate tokens.

In your MCP client​

The connection details page shows the exact config to add to your client. For Claude Code, this goes in .mcp.json. For Cursor, in MCP settings.

Once connected:

  • Prompts Server connections appear in your client's prompt palette (where supported). Select the prompt, fill in variables, and your client sends the rendered message.
  • Execution Server connections appear as callable tools. Your client can invoke them during a conversation and the result is returned inline.

Execution logs​

Every call through either server is logged. Open the prompt or workflow and go to Logs to see what was called, what inputs were passed, what was returned, and token usage. The source column shows mcp for these entries.

Access​

Connections are per-user. Each person who wants to use a prompt or workflow from their own MCP client sets up their own connection. Tokens can be revoked from the connection detail page at any time.