Repository navigation
MCP Development Guide
John Williams edited this page Mar 10, 2026
·
1 revision
How to build MCP (Model Context Protocol) servers for advertising platforms.
MCP is a protocol that lets AI tools (Claude, Cursor, Gemini) call external APIs through a standardized interface. An MCP server for Google Ads, for example, lets Claude pull live campaign data, run reports, and make changes β all within a conversation.
Every platform has an API. But connecting an AI tool to that API requires:
- Authentication handling
- Request formatting
- Response parsing
- Error handling
- Rate limiting
An MCP server wraps all of this into tools that AI can call naturally.
- Choose a platform β Check mcp-servers/registry.yaml for what's available
- Start from a template β Python template or Node template
- Implement the standard tools β See SPEC.md
-
Use core/ for auth β
core/auth/<platform>.pyhandles credentials -
Test with Claude Code β Add to your
claude_desktop_config.json - Submit a PR β Add to the registry and share with the community
Every advertising MCP server should implement at minimum:
-
list_campaignsβ List campaigns with status and metrics -
get_campaignβ Detailed campaign data -
get_metricsβ Performance metrics for date ranges -
list_audiencesβ Available audience segments -
get_budgetβ Budget and pacing status
See SPEC.md for full specifications.
The google-ads-mcp server is the reference implementation. Study its architecture for patterns you can apply to other platforms.
- Google Ads β‘ MCP Live
- Meta Ads
- Microsoft Ads
- Amazon Ads
- LinkedIn Ads
- Pinterest Ads
- Reddit Ads
- Spotify Ads
- The Trade Desk
- Demandbase
- Criteo
- AdRoll
- Quora Ads
- Agent Guide
- Paid Media Division β‘
- Platform Specialists
- Cross-Platform Agents
- Buddy
- Building Custom Agents
- Authentication Patterns
- Conversion Tracking
- Audience Strategy
- Reporting APIs
- Rate Limiting
- Budget Allocation
- Claude Code Β· Cursor Β· Gemini CLI
- Windsurf Β· Aider Β· OpenCode