Connect MCP and CLI clients - AutoAdy Help Center

Use MCP for clients that discover tools and the CLI for shell commands. Their credentials carry your AutoAdy access; setup is complete only after a verified tool result.

MCP server

The remote endpoint is:

https://www.autoady.io/api/mcp

OAuth-capable clients use account sign-in through OAuth 2.1. Follow Connect your own agent for the built-in recipes and Claude setup. Sign in with the intended identity and verify list_ad_accounts before allowing management tools. For clients that launch a local stdio server, use the published wrapper with a private API-key environment:

{
  "mcpServers": {
    "autoady": {
      "command": "npx",
      "args": ["-y", "@autoady/mcp-server"],
      "env": { "AUTOADY_API_KEY": "adk_YOUR_KEY" }
    }
  }
}

Replace the placeholder privately and follow the client’s configuration-file instructions. Do not commit a file containing the real key. MCP offers a wider tool catalog than REST. Discover it with tools/list and follow each returned input schema. Tool failures can be JSON-RPC results with isError: true, even when the transport request succeeded. Inspect content and any structuredContent; do not assume every result contains a structured block. Raw numeric reporting data, when supplied, carries its currency and applied window. For recurring reporting, check get_connected_account before interpreting old performance as proof of current delivery. Use explicit account_id and the date arguments accepted by each tool. Configure schedules in your own runner, or use AutoAdy scheduled tasks for supported read/report work. Scheduling does not unlock blocked ad writes.

Command line

The CLI requires Node.js 18 or later. Store AUTOADY_API_KEY in your private environment, then run:

npx -y @autoady/cli accounts

For a persistent local command:

npm install -g @autoady/cli
autoady results --account act_1234567890 --date-preset last_7d

Use a returned account ID in place of the example. A --key flag exists, but an environment credential avoids placing the secret in command history and process arguments.

Command	Result
`accounts`	Reachable Meta accounts
`results`	Performance for a reporting window
`health`	Account health and components
`benchmarks`	Industry benchmark data
`compare`	Your metrics compared with a benchmark
`fatigue`	A fatigue calculation from supplied numbers
`fatigue-alerts`	Account creative-fatigue observations
`ad-copy`, `hooks`	Paid-plan text generation
`call`	A tool call with a JSON object

Use --help for the installed version’s flags. For raw input, --json-body supplies the request; --json controls the output format:

autoady call get-skill --json-body '{"slug":"audit"}' --json
autoady call winner-insights --json-body '{"account_id":"act_1234567890"}' --json

The installed CLI has a known-tool manifest. If a newly published REST endpoint is rejected as an unknown tool, check the package version or call that endpoint directly over REST.

Exit codes and automation

Code	Meaning
0	The CLI accepted the HTTP response. Inspect the returned operation state, particularly HTTP 207 partial writes.
1	API rejection, request limit, network failure, or unexpected response.
2	Usage or authentication error, such as an unknown command, invalid flag, or missing/invalid key.

Output goes to stdout; diagnostics go to stderr. --json returns the response envelope for structured processing. A zero exit code alone is insufficient evidence that every step of a write completed. Follow response and retry handling before automating mutations.

​MCP server

​Command line

​Exit codes and automation

MCP server

Command line

Exit codes and automation