Meta Ads and Google Ads in Claude Code

Claude Code works where your files are, so the ad account joins the codebase. Ask for last week's performance_period at the ad level and have the figures written to a CSV, compare two periods with metric_compare before a budget review, or read campaigns_structure while you edit the tracking that feeds the account. Every result carries its exact data in structuredContent.result, the field to read in code.

Launches and changes work the same as in every client. launch_brief says what a clean launch still needs, campaigns_launch or google_ads_build prepares the campaign, and a pause, a budget change of at most 30%, an edit, a duplicate or a split test is prepared as one exact change you apply in Adrails.

• Pull figures into files, scripts and reports in your project
• Check campaign structure next to the code that sends conversions
• Prepare launches and changes, applied from Adrails

Claude Code adds remote MCP servers from the command line. By default the server is added for the current project only; --scope user adds it for every project, and --scope project writes it to a .mcp.json file at the project root that your team can share through version control. Claude Code asks before it uses a project-scoped server from .mcp.json.

If the browser does not open for the sign-in, copy the address Claude Code prints and open it yourself. Allowing a tool call in Claude Code is Claude Code's own setting; an advertising change waits on its page in Adrails.

• 1. From your project directory, run claude mcp add --transport http adrails https://adrails.ai/api/mcp
• 2. In Claude Code, run /mcp, select adrails and authenticate
• 3. Your browser opens Adrails: sign in, choose the workspace and the permissions, and choose Allow access
• 4. Return to Claude Code and ask your first question

The server registers 63 tools: the 43 tools the Adrails agents use in the app and 20 tools for accounts, integrations, campaign drafts, automations, Knowledge and job status. Claude Code lists only the tools the connection's permissions and your role allow. Reads return data and change nothing; a tool that changes an ad account prepares the change and needs an idempotency key, so a retried call never does the work twice.

Every tool, its inputs and who can call it is in the MCP tools reference, and the server card lists each tool with the schema the server registers, marked read-only or not.

The sign-in is OAuth 2.1 with dynamic client registration and PKCE: Claude Code registers itself, and no client ID, secret, API key or developer token is copied. Meta Ads and Google Ads are connected once in Adrails, which holds the platform access; tool results never contain provider credentials. The consent screen asks for one workspace and the permissions you grant, and a permission never goes beyond your role.

No tool creates, edits, pauses, switches on or rebudgets a campaign, ad set or ad by itself. A change comes back with a link to the Review advertising change page in Adrails, where an owner or admin applies it or declines it after Adrails reads the platform again. Everything a launch creates lands paused, switching it on is a second proposal that lists every budget that starts spending, and an applied change can be undone while the platform still matches what Adrails wrote. A proposal expires after 30 minutes.

Every tool call is recorded with who made it, the workspace, the client, the tool, the outcome and the duration, never the argument values. Each connection has its own history in the workspace. Revoke Claude Code at any time under Your connections on the MCP page in Adrails; removing the server with claude mcp remove does not revoke it there.