Connect an AI agent to Cover

Cover provides an MCP server for compatible AI agents. It gives the signed-in restaurant owner access to approved analytics and, when separately authorized, cost management tools.

Server and sign-in

Use this remote MCP URL in an MCP-compatible client:

https://mcp.coveranalytics.app/mcp

The server uses MCP over Streamable HTTP and OAuth Authorization Code with PKCE S256. The client discovers OAuth settings and supported scopes from the live server metadata. Sign in to Cover in the browser and approve only the access the client needs. Access is tied to the authenticated restaurant.

For an MCP client that accepts remote HTTP servers in this format, add:

{
  "servers": {
    "cover": {
      "type": "http",
      "url": "https://mcp.coveranalytics.app/mcp"
    }
  }
}

Start or reload the server in your client. When it connects, sign in to Cover and approve the scopes required for your task.

If your Cover account does not have MCP access, request access from Cover. A request does not itself grant access.

Choose scopes

The client should request the smallest scope set that supports the task. mcp:read is the baseline for analytics tools. The live protected-resource metadata is authoritative: never request a scope it does not advertise. The other scopes are optional where advertised:

Scope	Access
mcp:read	List approved analytics tables, inspect their definitions, and get query guidance. Also required for the separately scoped write tools.
mcp:sql:read	Run approved read-only SQL with execute_sql_readonly. Queries must be a single SELECT or WITH query.
mcp:food-costs:write	Add, correct or delete Menu Item food cost estimates.
mcp:employee-costs:write	Where advertised by the live server, add, correct or delete employee hourly cost rates.

Granting mcp:read alone does not grant SQL or write access. Each write scope is separate. Ask the restaurant owner before requesting or using a write scope, and get clear approval for each intended change. The agent must use amounts and dates supplied or confirmed by the owner; it must not infer food costs or wages.

Available tools

• Analytics: list_tables, get_table_info, get_query_guidance.
• SQL: execute_sql_readonly (mcp:sql:read).
• Food costs: find_menu_items, get_food_cost_history, add_food_cost_estimates, correct_food_cost_estimate, delete_food_cost_estimate (mcp:food-costs:write for changes).
• Employee rates, when deployed: find_employees, get_employee_cost_history, add_employee_cost_rates, correct_employee_cost_rate, delete_employee_cost_rate (mcp:employee-costs:write for changes).

Availability depends on the connected server. Use tools/list to confirm which tools exist, and require the live protected-resource metadata to advertise mcp:employee-costs:write before requesting or using employee write access. If the scope or tools are absent, treat employee-rate management as unavailable.

Food cost changes support previews, revision checks and idempotent retries. When employee-rate tools are available, their changes support the same safeguards. Read history before a correction or deletion and use its current revision. Employee rate tools change hourly rates only; they do not change restaurant-wide labor settings or extra-hours rules.

Discovery and trust

Use the live metadata to confirm the resource identifier and supported OAuth settings:

• Protected resource: https://mcp.coveranalytics.app/.well-known/oauth-protected-resource/mcp
• Authorization server: https://mcp.coveranalytics.app/.well-known/oauth-authorization-server

Read Cover’s security overview and privacy notice before connecting. The MCP service limits access to the authenticated restaurant and approved operations. It does not provide access to the POS for creating or editing sales.