> ## Documentation Index
> Fetch the complete documentation index at: https://docs.adsmcp.si/llms.txt
> Use this file to discover all available pages before exploring further.

# Run the ADSMCP Marketing Agent in Claude Code

> How the ADSMCP performance marketing agent runs inside Claude Code: subagent architecture, skills, slash commands, memory, and the agent loop.

Your AI advertising manager inside the terminal. The performance marketing agent runs as a Claude Code subagent — it has its own prompt, memory, and tools, and returns results back to your main session.

## How It Works

Claude Code uses an **agentic loop**: you describe a task, it reasons about what to do, calls tools, reads results, reasons again, and repeats until done. When your task involves advertising, Claude Code spawns the ADSMCP agent to handle it.

```mermaid theme={"theme":{"light":"github-light","dark":"github-dark"}} theme={null}
flowchart LR
    A[You] -->|"How are my Google Ads doing?"| B[Claude Code]
    B -->|Spawns subagent| C[Performance Marketing Agent]
    C -->|Reads| D[CLAUDE.md + STRATEGY.md + MEMORY.md]
    C -->|Calls| E[ADSMCP MCP Tools]
    E -->|OAuth| F[Google / Meta / LinkedIn / TikTok]
    C -->|Returns results| B
    B -->|Presents to you| A
```

### What Happens When You Ask About Ads

You say: *"Create a Google Search campaign for my SaaS product"*

1. Claude Code recognizes this as an advertising task
2. Spawns the performance marketing agent as a subagent
3. Agent reads `CLAUDE.md` (brand context), `STRATEGY.md` (directives), `MEMORY.md` (past decisions)
4. Follows the skill workflow: connection check → competitive research → keyword research → asset validation → campaign creation → ad extensions → verification
5. Each step calls ADSMCP MCP tools (`research_keywords`, `create_search_campaign`, etc.)
6. Agent asks for your confirmation before spending
7. Results return to your main Claude Code session

> [!TIP]
>
> The agent runs as a subagent with its own context — long advertising workflows don't bloat your main session.

***

## What Gets Installed

The ADSMCP plugin gives Claude Code everything it needs:

| Component | File | Purpose |
| - | - | - |
| **MCP Server** | `https://mcp.adsmcp.si/mcp` | 400+ advertising tools via OAuth |
| **Agent** | `~/.claude/agents/performance-marketing-agent.md` | Agent prompt with brand awareness and workflows |
| **Skill** | `~/.claude/skills/ad-campaign-management/SKILL.md` | Proven workflows for all platforms |
| **Commands** | `~/.claude/commands/*.md` | 5 slash commands for common tasks |

```bash theme={"theme":{"light":"github-light","dark":"github-dark"}} theme={null}
/plugin install adsmcp-ads-agent@claude-community
```

Available on the official Anthropic marketplace — no custom marketplace setup needed. Or type `/plugin`, open the **Discover** tab, and search `adsmcp`. Installs everything in one step. Run `/reload-plugins` after installing.

***

## Skills & Slash Commands

Skills are instruction files that teach Claude Code the *right* workflow for using ADSMCP's 400+ tools. Without skills, the AI guesses which tools to call and in what order. With skills, it follows proven advertising workflows with safety rules baked in.

### How Skills Load

1. **Skill descriptions** are always in context — Claude Code knows what's available
2. **Full skill content** loads when invoked (by you or automatically)
3. **The agent preloads the skill** when spawned, so it starts with complete workflow instructions

### One Comprehensive Skill

Claude Code uses **1 skill** (`ad-campaign-management`) covering all platforms and workflows. This differs from Cursor and Codex which split into 5 separate skills — Claude Code consolidates because its plugin system handles slash commands separately.

The skill includes:

* Workflow routing: maps your request to the right tool sequence
* Platform-specific campaign creation flows (Google, Meta, Amazon, ChatGPT Ads, LinkedIn, TikTok)
* Keyword research with strategy directive filtering
* Cross-platform performance analysis
* Budget optimization and wasted spend detection
* Ad extension workflows (sitelinks, callouts, structured snippets)
* Competitive intelligence using web research
* Safety rules: confirmation gates, PAUSED creation, read-before-write

### Slash Commands

| Command | What It Does |
| - | - |
| `/adsmcp:setup` | Bootstrap a brand workspace — connect accounts, scan docs, pull data |
| `/adsmcp:ad-campaign-management` | Full campaign management — all platforms, all workflows |
| `/adsmcp:performance-review` | Cross-platform performance scorecard |
| `/adsmcp:write-ad-copy` | Brand-voice ad copy from real performance data |
| `/adsmcp:wasted-spend` | Find and fix wasted ad spend |

You don't need to memorize these. Just describe what you want — Claude Code matches the right skill automatically.

Cross-platform performance review with strategy alignment.

Run a performance review for the last 30 days across all platforms. Compare against my KPI targets and flag any campaigns that conflict with my strategy.

> **[Sign up for ADSMCP — free to start](https://adsmcp.si)**
>
> 15 free tool calls/month. No credit card required. Connect your ad accounts in 2 minutes.

***

## Context Files

The agent uses three persistent files to maintain brand awareness across sessions.

### CLAUDE.md — Brand Context

Created by `/adsmcp:setup`. Claude Code reads this automatically at session start.

| Section | What It Contains |
| - | - |
| Brand Overview | What you sell, who you sell to, industry |
| Brand Voice | Tone, language style, prohibited words |
| Target Audiences | Segments with platform-specific targeting |
| Active Platforms | Connected platforms and campaign counts |
| Budget & Guardrails | Monthly budget, CPC caps, CPA targets, ROAS minimums |
| KPI Targets | Primary goals and target metrics |
| Performance Snapshot | Last 30 days across all platforms |
| Key Findings | Top campaigns, wasted spend, opportunities |

### STRATEGY.md — Strategic Decisions

Persists directives across sessions. Every skill reads this before executing.

```markdown theme={"theme":{"light":"github-light","dark":"github-dark"}} theme={null}
### Google Ads
AVOID: broad match "plumbing services" — competitor-dominated, $12+ CPC
PREFER: exact match "emergency plumber [city]" — high intent, $4-6 CPC
CONSTRAINT: max CPC $8 for non-brand terms — budget protection
```

Directives are **only saved after you explicitly confirm them.** The agent proposes, you approve.

### MEMORY.md — Past Decisions

Tracks campaign actions, optimization results, your preferences, and key learnings. The agent reads memory at session start and cross-references it during analysis.

***

## The Agent Loop: Step by Step

Here's exactly what happens when you run `/adsmcp:performance-review`:

**1. Agent spawns**

Claude Code creates a subagent with the performance marketing agent prompt and skill preloaded.

**2. Context loads**

Agent reads `CLAUDE.md`, `STRATEGY.md`, and `MEMORY.md`.

**3. Connections check**

Calls `get_connections_status` to identify which ad platforms are active.

**4. Data pull**

For each connected platform, calls performance and wasted spend tools in parallel.

**5. Strategy check**

Compares campaign data against `STRATEGY.md` directives. Flags "Strategy Drift" items.

**6. Analysis & recommendations**

Compares actuals vs KPI targets. Cross-references with `MEMORY.md`. Presents a unified scorecard with top 3 recommended actions.

***

## Web Research

The Claude Code agent has access to `WebSearch` and `WebFetch` — it can research competitors, crawl websites, and gather market intelligence. This powers the **Campaign Research** workflow that runs before any campaign creation:

1. `WebFetch` crawls your website and competitor sites for positioning, pricing, value propositions
2. `WebSearch` finds competitors, comparison content, and market data
3. Combined with ADSMCP data (search terms, keyword volumes), the agent creates a research brief before building your campaign

The agent doesn't just create campaigns from your prompt — it researches your market first.

***

## Safety Rules

| Rule | How It Works |
| - | - |
| User confirmation for spend | Agent always asks before creating campaigns or changing budgets |
| Campaigns created PAUSED | All `create_*` tools default to PAUSED status |
| Read-before-write | Connection check → research → validate → create |
| Never retry on error | Reports the error instead of retrying campaign creation |
| Budget guardrails | Checks `CLAUDE.md` budget limits before spend-affecting actions |
| Strategy compliance | Reads `STRATEGY.md` and flags conflicts with active directives |
| Post-creation verification | Verifies ad groups, keywords, ads, and extensions after creation |

***

## Comparison with Other Clients

| Feature | Claude Code | Cursor | Codex |
| - | :-: | :-: | :-: |
| Agent type | Subagent | Subagent | Agent config |
| Brand context file | `CLAUDE.md` | `BRAND.md` | `AGENTS.md` |
| Skills | 1 comprehensive | 5 separate | 5 separate |
| Slash commands | 5 (`/adsmcp:*`) | 5 (`/adsmcp-*`) | 5 (`$adsmcp-*`) |
| Memory | `MEMORY.md` | `MEMORY.md` | Not available |
| Rules | -- | 2 rule files | Safety rules |
| Web research | `WebSearch` + `WebFetch` | `WebSearch` + `WebFetch` | Not available |

***

## FAQ

> **How is the agent different from just using MCP tools?**
>
> Without the agent, you have 400+ tools but no workflow guidance, no brand awareness, no strategy persistence, and no memory. The agent turns raw tools into a managed advertising system that knows your brand and follows proven workflows.

> **Can I customize the agent prompt?**
>
> Yes. Edit `~/.claude/agents/performance-marketing-agent.md` to change behaviors, add workflows, or modify interaction style. The source is in the [ads-mcp repo](https://github.com/amekala/ads-mcp/tree/main/agents).

> **Does the agent use my main context window?**
>
> No. The agent runs as a subagent with its own context. Results are summarized and returned. Long advertising workflows don't bloat your main session.

> **Can I run the agent in the background?**
>
> Yes. Claude Code supports background agents. Spawn the agent for a long task (like a full cross-platform review) and continue working while it runs.

> **What model does the agent use?**
>
> The agent uses the same model as your Claude Code session. You can override this in the agent file with the `model` field.

## Related Documentation

* [Cursor Agent](https://docs.adsmcp.si/agent-skills/cursor-agent.md) — How ADSMCP works in Cursor
* [Codex Agent](https://docs.adsmcp.si/agent-skills/codex-agent.md) — How ADSMCP works in Codex
* [Performance Marketing Agent](https://docs.adsmcp.si/agent-skills/agent.md) — Architecture overview
* [Skill Reference](https://docs.adsmcp.si/agent-skills/skills.md) — All 5 skills with invocation details
* [Claude Code Setup](https://docs.adsmcp.si/ai-clients/claude-code.md) — Installation guide
* [Prompt Engineering Playbook](https://docs.adsmcp.si/agent-skills/prompt-engineering-playbook.md) — A custom skill, account dossier, and prompt book generated from your own ad account (\$39 one-time, one free on annual)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.