Pick the question you have.
Every page states the limit as well as the feature. If a number is not in the docs, it is not a promise.
Install AdRails with Claude Code
Install from GitHub, write one config file, and register the local stdio server with claude mcp add adrails -- python3 -m adsops --config /abs/path/config.json. There is no AdRails account and nothing to sign up for.
What the guardrails do, and don't, prevent
AdRails refuses budgets outside your band, reused or expired plan IDs, spend increases while the guard heartbeat is stale, and any tool write to the cap file — and it does not bound what Google delivers.
/docs/guardrails/ 03 · SIDE BY SIDERun beside Google's read-only Ads MCP
Google's official google-ads-mcp answers ad-hoc GAQL exploration and never writes; AdRails is the write path, and the two run side by side as one stdio entry each in the same client.
Let an AI agent manage Google Ads safely
An agent can manage Google Ads safely only when every write is a reviewed file change, caps live where the agent's tools cannot write, and a scheduled check pauses spend when nobody is watching.
/docs/ai-agent-google-ads/ 05 · CLIENTSGrok and other MCP clients
AdRails speaks JSON-RPC 2.0 over stdio, so any client that can launch python3 -m adsops --config /abs/path/config.json can run it; the page lists the generic mcpServers shape and the Claude Code command.
AdRails vs read-only and generic write servers
AdRails is the write path beside Google's read-only MCP and takes changes only through a reviewed desired-state file with policy-file caps; generic write servers differ vendor to vendor, so check each rail yourself.
/compare/pipx install git+https://github.com/antonioac1/adrails (the PyPI release is coming soon) — then run python3 -m adsops --selftest offline. Leave commit_policy at its default of allow_commit: false until you have read a plan and watched it apply.
Apache-2.0 · local stdio · your credentials, your spend