Skip to main content
Glama
arcticgreyy

Paid Media MCP

by arcticgreyy
OverviewSchemaRelated ServersScoreDiscussions
TypeScript
Local

Paid Media MCP

A Model Context Protocol (MCP) server template for paid media teams. Connect Claude to your campaign data, team structure, performance history, attribution models, reporting templates, asset library, test-and-learn history, audience library, and measurement setup — so it can answer questions, write reports, debug tracking issues, and assist with analysis directly in your workflow.

This is a template. The example data is for a fictional e-commerce company ("Acme Corp"). You replace it with your own.

Table of Contents

  1. What this does

  2. Quick start

  3. Filling in your data

  4. What Claude can do

  5. Example conversations

  6. Connecting to live APIs

  7. Troubleshooting

  8. Project structure

What this does

When connected to Claude, this MCP server gives Claude real knowledge of your paid media operation:

  • Your campaigns — platforms, budgets, objectives, targeting, funnel stage, status

  • Your team structure — who manages what, their KPIs, platforms, and responsibilities

  • Your performance data — historical metrics you've loaded, with automatic aggregation

  • Your attribution setup — how you measure conversions, window settings, model comparisons

  • Your reporting formats — templates for weekly team reports, executive decks, pacing dashboards

Without this, Claude can only give generic paid media advice. With it, Claude can say:

"Looking at your Q2 Meta prospecting campaign, it's running a 2.4x ROAS against your 4.0x target. The retargeting campaign is at 12.3x, which suggests the prospecting pool may be too cold — worth tightening the lookalike to 1% and suppressing recent purchasers."

Quick start

Prerequisites

Step 1 — Clone and install

git clone https://github.com/arcticgreyy/paidmedia-mcp.git
cd paidmedia-mcp
npm install

Step 2 — Fill in your data

Edit the JSON files in the data/ folder. Each file has detailed examples. See the Filling in your data section for field-by-field guidance.

At minimum, fill in:

  • data/metadata.json — your company name

  • data/accounts.json — your ad accounts

  • data/teams.json — your team(s)

The server will start even if other files are missing or partially filled.

Step 3 — Build

npm run build

For development without building (requires tsx):

npm run dev

Step 4 — Connect to Claude Desktop

Open your Claude Desktop config file:

  • Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following (replace the paths with your actual absolute paths):

{
  "mcpServers": {
    "paid-media": {
      "command": "node",
      "args": ["/absolute/path/to/paidmedia-mcp/dist/index.js"],
      "env": {
        "PAID_MEDIA_DATA_DIR": "/absolute/path/to/paidmedia-mcp/data"
      }
    }
  }
}

To find your absolute path, run this from the project folder:

pwd

Copy the output and append /dist/index.js for the args value and /data for PAID_MEDIA_DATA_DIR.

Step 5 — Restart Claude Desktop

Fully quit and reopen Claude Desktop. You should see a hammer icon (🔨) in the chat input — click it to verify the paid-media tools are listed.

Updating data

After editing files in data/, restart Claude Desktop (or the MCP server process) for changes to take effect. The server loads data files at startup.

Filling in your data

metadata.json

Basic company info. Used in report headers and resource descriptions.

{
  "metadata": {
    "company_name": "Your Company Name",
    "industry": "E-commerce",
    "primary_currency": "USD",
    "fiscal_year_start": "01-01",
    "last_updated": "2026-05-31"
  }
}

Field

Required

Notes

company_name

Yes

Appears in Claude's context

industry

No

Helps Claude give relevant benchmarks and advice

primary_currency

Yes

ISO 4217 code (USD, EUR, GBP, etc.)

fiscal_year_start

Yes

MM-DD format

last_updated

Yes

ISO date — update when you refresh data

accounts.json

One entry per ad account. Each account belongs to exactly one team.

{
  "accounts": [
    {
      "id": "acc_brand_gads",
      "name": "Brand — Google Ads",
      "platform": "google_ads",
      "platform_account_id": "123-456-7890",
      "team_id": "team_brand",
      "status": "active",
      "currency": "USD",
      "timezone": "America/New_York",
      "monthly_budget": 50000,
      "notes": "Brand campaigns only"
    }
  ]
}

Field

Required

Notes

id

Yes

Your internal ID — used in campaigns.json and teams.json

name

Yes

Human-readable name for Claude to display

platform

Yes

See platform values below

platform_account_id

Yes

The actual ID shown in the platform UI (e.g. Google Ads customer ID)

team_id

Yes

Must match an id in teams.json

status

Yes

active or inactive

currency

Yes

ISO 4217

timezone

Yes

IANA timezone string

monthly_budget

No

Used for pacing analysis

notes

No

Free text — Claude reads this

Platform values: google_ads · meta · dv360 · youtube · linkedin · tiktok · twitter_x · pinterest · snapchat · amazon · other

teams.json

One entry per media team. This is where you capture the strategic context Claude needs to give relevant advice.

{
  "teams": [
    {
      "id": "team_brand",
      "name": "Brand & Awareness",
      "description": "Manages upper-funnel brand campaigns focused on reach and video.",
      "objectives": ["awareness", "reach", "video_views"],
      "primary_kpis": ["impressions", "reach", "cpm", "brand_lift"],
      "account_ids": ["acc_brand_gads", "acc_brand_dv360"],
      "member_ids": ["user_sarah", "user_james"],
      "lead_id": "user_sarah",
      "platforms": ["google_ads", "dv360", "youtube"],
      "reporting_cadence": "weekly",
      "budget_owner": "vp_marketing",
      "notes": "Brand lift studies run quarterly. All campaigns follow brand safety guidelines."
    }
  ]
}

Field

Required

Notes

id

Yes

Referenced by accounts and campaigns

objectives

Yes

List from: awareness reach traffic engagement video_views lead_generation app_installs conversions catalog_sales store_visits

primary_kpis

Yes

Free-text metric names — Claude uses these when reviewing performance

account_ids

Yes

Must match IDs in accounts.json

member_ids

Yes

Must match IDs in team-members.json

lead_id

Yes

Team lead — must be in member_ids

reporting_cadence

Yes

daily weekly biweekly monthly

notes

No

Put key context here: targets, processes, constraints

Tip: The notes and description fields are the most valuable for Claude. Include:

  • ROAS/CPA/CPL targets

  • Key rules or constraints (e.g. "brand safe only")

  • Important processes (e.g. "budget reviews every Monday")

  • Links or references to external docs

team-members.json

One entry per person. Members can belong to multiple teams.

{
  "team_members": [
    {
      "id": "user_mike",
      "name": "Mike Rodriguez",
      "email": "mike@yourcompany.com",
      "role": "director",
      "team_ids": ["team_performance"],
      "platform_specialties": ["google_ads", "meta"],
      "responsibilities": [
        "Performance team leadership",
        "ROAS and CPA target setting",
        "Budget allocation across channels"
      ],
      "notes": "Primary stakeholder for attribution model decisions."
    }
  ]
}

Field

Required

Notes

id

Yes

Must match member_ids in teams.json

role

Yes

media_buyer media_planner analyst strategist manager director other

team_ids

Yes

Array — analysts or shared staff can be on multiple teams

platform_specialties

Yes

Which platforms they work in day-to-day

responsibilities

Yes

Free-text list — Claude uses this to route questions to the right person

notes

No

Useful for capturing context like certifications, agency relationships, or ownership areas

campaigns.json

One entry per campaign. This is the largest and most important data file.

{
  "campaigns": [
    {
      "id": "camp_perf_search_001",
      "name": "Performance Search — Always On",
      "platform": "google_ads",
      "account_id": "acc_perf_gads",
      "team_id": "team_performance",
      "status": "active",
      "objective": "conversions",
      "budget": {
        "type": "daily",
        "amount": 2000,
        "currency": "USD",
        "pacing": "standard"
      },
      "start_date": "2026-01-01",
      "end_date": null,
      "targeting": {
        "geo": ["US"],
        "devices": ["desktop", "mobile"],
        "keyword_themes": ["branded", "competitor", "category"]
      },
      "funnel_stage": "lower",
      "tags": ["always-on", "search", "performance"],
      "notes": "Target ROAS: 5.0x. Smart Bidding tROAS strategy."
    }
  ]
}

Budget types: daily · lifetime · monthly

Objective values: awareness · reach · traffic · engagement · video_views · lead_generation · app_installs · conversions · catalog_sales · store_visits

Funnel stages: upper · mid · lower

Status values: active · paused · ended · draft · archived

Targeting fields (all optional):

Field

Example

geo

["US", "CA", "GB"]

audience_segments

["lookalike_1pct_purchasers", "in-market_home_goods"]

age_range

{ "min": 25, "max": 54 }

gender

"all" · "male" · "female"

devices

["desktop", "mobile", "tablet", "connected_tv"]

keyword_themes

["branded", "competitor", "generic"]

custom_audiences

["cart_abandoners_14d", "crm_suppression"]

Tags tip: Tags are searchable. Use them consistently for filtering — e.g. q2-2026, always-on, brand, retargeting, test.

historical-performance.json

Contains two sections: daily records and benchmarks.

Records

One entry per campaign per day. Include whatever metrics you have — rate metrics are re-calculated automatically when aggregating.

{
  "records": [
    {
      "campaign_id": "camp_perf_search_001",
      "date": "2026-05-01",
      "metrics": {
        "impressions": 125000,
        "clicks": 4200,
        "spend": 1850.00,
        "conversions": 310,
        "conversion_value": 12400.00
      }
    }
  ]
}

Available metric fields:

Metric

Description

impressions

Total ad impressions

clicks

Total clicks

spend

Total spend in account currency

reach

Unique users reached

frequency

Average impressions per user

ctr

Click-through rate (%)

cpc

Cost per click

cpm

Cost per 1,000 impressions

cpa

Cost per acquisition/conversion

roas

Return on ad spend

conversions

Total conversion events

conversion_value

Revenue value of conversions

video_views

Video view count

video_view_rate

Percentage who watched

view_through_conversions

View-assisted conversions

custom_metrics

Any additional metrics as key-value pairs

Tip: You don't need to pre-calculate rate metrics (CTR, CPC, ROAS). The server calculates them when aggregating. Just include the raw counts.

How to export data: Most platforms let you download daily campaign performance as CSV. Use a tool like Google Sheets or Python to convert to this JSON format. See data/historical-performance.json for the full structure.

Benchmarks

Industry reference points for comparison. Optional but makes performance analysis much richer.

{
  "benchmarks": [
    {
      "platform": "google_ads",
      "objective": "conversions",
      "industry": "e-commerce",
      "avg_ctr": 2.69,
      "avg_cpc": 1.33,
      "avg_cpa": 33.52,
      "avg_roas": 3.5
    }
  ]
}

Use published industry benchmarks from WordStream, Meta, Google, or your agency. Update annually.

attribution-models.json

Documents how you measure conversions — not connected to live platforms, purely descriptive. Claude uses this to explain your measurement setup and make recommendations.

{
  "attribution_configurations": [
    {
      "id": "attr_last_click_30d",
      "name": "Last Click — 30-Day",
      "model": "last_click",
      "window": {
        "click": 30,
        "view": 0,
        "unit": "days"
      },
      "platforms_applied": ["google_ads", "meta"],
      "conversion_events": ["purchase", "lead_form_submit"],
      "cross_device": false,
      "view_through_enabled": false,
      "description": "100% credit to the last clicked ad within 30 days.",
      "use_cases": [
        "Bottom-funnel campaign optimization",
        "Simple baseline for cross-channel comparison"
      ],
      "notes": "Over-credits search/retargeting; undervalues prospecting."
    }
  ]
}

Model values: last_click · first_click · linear · time_decay · position_based · data_driven · custom

Tip: Add one entry per meaningful configuration you use — e.g. one for Google Ads, one for Meta, one for cross-channel analysis. The use_cases and notes fields are especially valuable for Claude's analysis.

reporting-templates.json

Defines the structure of your reports and dashboards. Claude uses these when generating reports to follow the right format for the right audience.

{
  "reporting_templates": [
    {
      "id": "tmpl_weekly_team",
      "name": "Weekly Performance Report",
      "type": "performance_summary",
      "audience": "media_team",
      "cadence": "weekly",
      "metrics_included": ["spend", "conversions", "roas", "cpa"],
      "dimensions": ["campaign", "platform", "date"],
      "visualizations": ["spend_by_campaign_bar", "roas_trend_line"],
      "delivery_format": ["google_sheets", "slack"],
      "sections": [
        {
          "title": "Week Summary",
          "description": "Overall performance vs. prior week and vs. targets.",
          "metrics": ["spend", "conversions", "roas"]
        },
        {
          "title": "Campaign Scorecards",
          "description": "One row per active campaign with pacing status.",
          "metrics": ["spend", "conversions", "cpa", "roas"]
        },
        {
          "title": "Action Items",
          "description": "Specific actions to take next week.",
          "metrics": []
        }
      ]
    }
  ]
}

Audience values: executive · media_team · client · internal

Type values: performance_summary · pacing · budget_flight · audience_insights · creative_analysis · channel_mix · attribution_path · executive_summary · custom

Tip: The sections array is the most important part — it's the outline Claude follows when writing a report. Make section descriptions as specific as possible.

assets.json

Documents your creative asset library and per-platform specs. Claude uses this to answer asset questions, check spec compliance, and guide campaign trafficking.

{
  "asset_library": {
    "dam_system": "Bynder",
    "dam_url": "https://yourcompany.bynder.com",
    "access_instructions": "Request access via IT helpdesk. Login with SSO.",
    "brand_guidelines_url": "https://yourcompany.bynder.com/brand-guidelines",
    "copy_guidelines_url": "https://yourcompany.bynder.com/copy-guidelines",
    "categories": [
      {
        "id": "cat_social_video",
        "name": "Social Video",
        "type": "video",
        "location_url": "https://yourcompany.bynder.com/collections/social-video",
        "naming_convention": "{brand}_{campaign}_{duration}s_{platform}_{version}",
        "specs": [
          {
            "platform": "meta",
            "format": "MP4 or MOV",
            "dimensions": "1080x1080, 1080x1920, 1920x1080",
            "max_file_size": "4GB",
            "duration_max": "60s (Reels), 15s (Stories)",
            "notes": "Include captions — 85% watch without sound"
          }
        ]
      }
    ]
  }
}

Asset type values: image · video · copy · audio · html5 · document · other

Tip: The specs array per category is especially valuable — Claude can answer "what do I need to traffic this campaign on TikTok?" directly from this data.

testing.json

Documents your entire test-and-learn program: methodology, tools, and every test (active, planned, completed). This gives Claude a searchable institutional memory of what your team has tested and learned.

{
  "testing_program": {
    "methodology": {
      "confidence_threshold": 95,
      "require_stat_sig": true,
      "minimum_sample_size": 1000,
      "minimum_test_duration_days": 14,
      "minimum_detectable_effect_pct": 10,
      "winner_criteria": "≥10% lift at 95% confidence, minimum 14 days runtime. If stat sig not achieved, test is inconclusive — retain prior best practice."
    },
    "tools": [
      {
        "id": "tool_meta_ab",
        "name": "Meta A/B Testing (native)",
        "type": "platform_native",
        "platform": "meta",
        "used_for": ["Creative A/B tests", "Audience split tests"]
      }
    ],
    "tests": [
      {
        "id": "test_001",
        "name": "UGC vs Brand Creative",
        "hypothesis": "UGC content will drive lower CPA on Meta prospecting due to higher thumb-stop rate.",
        "status": "completed",
        "type": "creative",
        "platform": "meta",
        "variants": [
          { "id": "v_control", "name": "Brand Creative", "description": "Polished brand video", "is_control": true },
          { "id": "v_ugc", "name": "UGC Video", "description": "Creator-style mobile video", "is_control": false }
        ],
        "results": {
          "winner_variant_id": "v_ugc",
          "stat_sig_achieved": true,
          "confidence_level": 97,
          "primary_metric": "CPA",
          "primary_metric_lift_pct": -23,
          "conclusion": "UGC drove 23% lower CPA.",
          "action_taken": "Scaled UGC — producing 2 new variants monthly."
        }
      }
    ]
  }
}

Field

Required

Notes

methodology.confidence_threshold

Yes

90 or 95 are most common

methodology.winner_criteria

Yes

Plain-English rule — Claude quotes this when analyzing tests

tests[].hypothesis

Yes

Claude uses this to evaluate whether the test was well-designed

tests[].results.primary_metric_lift_pct

No

Negative = improvement for cost metrics (CPA, CPC)

Test type values: creative · audience · bidding · landing_page · copy · format · offer · other

Test status values: planned · active · completed · paused · abandoned

Tip: Historical tests are the most valuable part — the more completed tests you document, the better Claude can summarize learnings and recommend next experiments.

audiences.json

Documents your full audience infrastructure: first-party segments, contracted data providers, data onboarding platforms, lookalike strategy, and third-party overlay layers.

{
  "audience_library": {
    "business_unit": "Consumer",
    "first_party_audiences": [
      {
        "id": "aud_purchasers_180d",
        "name": "Past Purchasers — 180 Days",
        "type": "pixel_based",
        "description": "All purchasers in the last 180 days.",
        "size_estimate": 45000,
        "platforms_available": ["meta", "google_ads"],
        "refresh_cadence": "daily",
        "tags": ["purchasers", "lookalike-seed"]
      }
    ],
    "data_providers": [
      {
        "id": "dp_oracle",
        "name": "Oracle Data Cloud",
        "contract_status": "active",
        "segments_available": ["In-Market: Home & Garden"],
        "platforms_available": ["dv360", "google_ads"],
        "contract_end_date": "2026-12-31"
      }
    ],
    "onboarding_platforms": [
      {
        "id": "ob_liveramp",
        "name": "LiveRamp",
        "type": "data_onboarding",
        "platforms_connected": ["meta", "google_ads", "dv360"],
        "use_cases": ["CRM list onboarding", "Identity resolution"]
      }
    ],
    "lookalike_strategy": {
      "default_expansion_pct": 1,
      "entries": [
        {
          "seed_audience_id": "aud_purchasers_180d",
          "seed_audience_name": "Past Purchasers — 180 Days",
          "platform": "meta",
          "expansion_percentages": [1, 2, 5],
          "best_performing_expansion": 1
        }
      ]
    },
    "third_party_layers": [
      {
        "id": "3p_in_market",
        "name": "In-Market: Your Category",
        "provider": "Google",
        "category": "in-market",
        "platforms_available": ["google_ads", "dv360"],
        "is_default": true,
        "is_best_performer": true,
        "cpm_premium_estimate": 0.50
      }
    ]
  }
}

First-party audience types: crm_list · pixel_based · customer_match · email_list · app_users · lookalike_seed · suppression · other

Onboarding platform types: clean_room · data_onboarding · identity_resolution · cdp

Tip: Set is_default: true on audience layers you apply to every campaign, and is_best_performer: true on layers with proven efficiency gains — Claude uses both flags to make targeting recommendations.

measurement.json

Documents your full tracking and measurement infrastructure: tag management, pixels, Conversion APIs, CM360/Floodlight setup, website data layer, and measurement partners.

{
  "measurement_setup": {
    "tag_management": {
      "system": "google_tag_manager",
      "container_id": "GTM-XXXXXXX",
      "implementation_type": "hybrid",
      "server_side_endpoint": "https://sst.yourcompany.com"
    },
    "pixels_and_tags": [
      {
        "id": "pixel_meta",
        "name": "Meta Pixel",
        "platform": "meta",
        "pixel_id": "XXXXXXXXXXXXXXXX",
        "implementation": "both",
        "events_tracked": ["PageView", "AddToCart", "Purchase"],
        "custom_parameters": ["value", "currency", "order_id"]
      }
    ],
    "conversion_apis": [
      {
        "id": "capi_meta",
        "platform": "meta",
        "api_name": "Meta Conversions API (CAPI)",
        "implementation": "both",
        "events_sent": ["Purchase", "Lead"],
        "match_rate_estimate_pct": 87,
        "deduplication_method": "event_id matched client-side and server-side"
      }
    ],
    "cm360": {
      "account_id": "XXXXXXX",
      "floodlight_configuration_id": "FL-XXXXXXXX",
      "u_variables": [
        { "variable": "u1", "name": "order_id", "type": "string", "description": "Unique order ID for deduplication" },
        { "variable": "u2", "name": "order_value", "type": "number", "description": "Order revenue in USD" }
      ]
    },
    "website_data_capture": {
      "data_layer_implemented": true,
      "data_layer_spec_url": "https://docs.yourcompany.com/datalayer",
      "data_layer_variables": ["pageType", "transactionId", "transactionRevenue", "userId"],
      "analytics_platform": "Google Analytics 4",
      "analytics_property_id": "G-XXXXXXXXXX",
      "first_party_cookies_implemented": true,
      "cookie_domain": ".yourcompany.com",
      "cookie_session_duration": "13 months",
      "cookie_data_captured": ["_ga", "_gcl_aw (GCLID)", "_fbc (Meta click ID)"]
    },
    "measurement_partners": [
      {
        "id": "mp_mmm",
        "name": "Your MMM Provider",
        "type": "mmm",
        "status": "active",
        "platforms_covered": ["google_ads", "meta", "dv360"],
        "cadence": "quarterly"
      }
    ]
  }
}

Tag management system values: google_tag_manager · tealium · adobe_launch · segment · mparticle · manual · other

Implementation type values: client_side · server_side · hybrid

Measurement partner types: mmm · incrementality · brand_lift · attribution · analytics · identity · other

Tip: The cm360.u_variables section is especially useful — Claude can explain what each u-variable captures and help troubleshoot Floodlight discrepancies when you reference specific variable names.

What Claude can do

Tools

Tools are actions Claude takes to retrieve your data.

Campaign tools

Tool

What it does

list_campaigns

Filter campaigns by team, platform, status, objective, funnel stage, or tag

get_campaign

Full details for one campaign by ID

list_accounts

All accounts, optionally filtered by team

get_account

Full account details by ID

Team tools

Tool

What it does

list_teams

All teams with objectives, KPIs, platforms

get_team

Team details + its members + its accounts

get_team_for_account

Which team owns a given account

list_team_members

Members, optionally filtered by team

get_team_member

Full member details

Performance tools

Tool

What it does

get_campaign_performance

Per-day records + aggregated totals for a campaign

get_team_performance

Aggregated totals by team, broken down per campaign

get_benchmarks

Industry benchmarks by platform and objective

Attribution tools

Tool

What it does

list_attribution_models

All configured attribution setups

get_attribution_model

Full details for one model

compare_attribution_models

Side-by-side diff of two models

Reporting tools

Tool

What it does

list_reporting_templates

Templates filtered by audience

get_reporting_template

Full template with all sections

build_performance_report

Assembles raw data + template for Claude to narrate

Asset tools

Tool

What it does

get_asset_library

DAM system info, access instructions, guidelines links, category summary

list_asset_categories

All asset categories, optionally filtered by type (image, video, copy, etc.)

get_asset_category

Full details for one category: location, naming convention, specs

get_asset_specs

Platform-specific specs (dimensions, file size, format) for a given asset type and platform

Testing tools

Tool

What it does

get_testing_methodology

Confidence threshold, stat sig rules, winner criteria, and testing tools in use

list_tests

All tests filtered by status, type, team, or platform

get_test

Full details for one test: hypothesis, all variants, results

get_test_learnings

Summarized learnings from all completed tests

Audience tools

Tool

What it does

get_audience_library_overview

High-level summary: 1P count, data providers, onboarding platforms, LAL strategy

list_first_party_audiences

All 1P segments filtered by business unit or platform

list_data_providers

Contracted data providers filtered by contract status

get_lookalike_strategy

Full LAL strategy: seed audiences, expansion sizes, best performers

list_third_party_audience_layers

Overlay segments filtered by platform, default use, or best-performer flag

get_onboarding_platforms

Data onboarding and clean room platforms in use

Measurement tools

Tool

What it does

get_measurement_overview

High-level tracking stack: TMS, pixel count, CAPI count, measurement partners

get_tag_management

TMS details: platform, container ID, implementation type, server-side endpoint

list_pixels_and_tags

All platform pixels filtered by platform: events tracked, implementation

list_conversion_apis

All Conversion API implementations: events, match rate, deduplication method

get_cm360_setup

CM360 account, Floodlight config ID, and all u-variables with descriptions

get_website_data_capture

Data layer status/variables, analytics platform, first-party cookie setup

list_measurement_partners

MMM, incrementality, brand lift, attribution partners filtered by type

Resources

Resources are documents Claude can read as context (similar to attaching a file).

Resource URI

What it contains

paid-media://overview

Company metadata + all teams + all accounts

paid-media://campaigns

All campaigns

paid-media://team-structure

Teams enriched with their members

paid-media://attribution-models

All attribution configurations

paid-media://reporting-templates

All reporting templates

paid-media://asset-library

Full asset library: DAM info, categories, and specs

paid-media://testing-program

Full testing program: methodology, tools, and all tests

paid-media://audience-library

Full audience library: 1P, providers, LAL strategy, 3P layers

paid-media://measurement-setup

Full tracking stack: TMS, pixels, APIs, CM360, data capture

Prompts

Pre-built prompt templates that guide Claude through multi-step analysis tasks. Access them from Claude's prompt picker (the / command in supported clients).

Prompt

Arguments

What it does

campaign_performance_review

campaign_id, date_from, date_to

Full performance review with pacing and benchmark comparison

team_weekly_report

team_id, week_start

Weekly scorecard using your team's reporting template

attribution_analysis

campaign_id

Explains model options and recommends the right one for the campaign objective

budget_pacing_check

team_id (optional), check_date

Flags over/under-pacing campaigns with specific actions

channel_mix_analysis

team_id, date_from, date_to

Channel efficiency ranking and reallocation recommendations

test_and_learn_review

team_id (optional), platform (optional)

Summarizes test history, active tests, gaps, and recommends next experiments

audience_strategy_review

platform

Full audience strategy audit: 1P usage, LAL quality, 3P layer gaps, recommendations

measurement_audit

(none)

End-to-end tracking audit: signal gaps, deduplication risks, attribution reliability

asset_readiness_check

campaign_id

Checks whether required assets exist for a campaign's platform and generates a trafficking checklist

Example conversations

Once connected, you can ask Claude things like:

Campaign lookup

"What active Meta campaigns does the performance team have running right now?"

"Show me all upper-funnel campaigns tagged q2-2026."

Performance analysis

"How did the search always-on campaign perform last week? Compare it to our benchmarks."

"Which of our campaigns had the best ROAS in May?"

Pacing

"Is our Q2 Meta prospecting campaign on pace to deliver its budget?"

"Check pacing across all active campaigns as of today."

Team and ownership

"Who manages the LinkedIn account and what are their KPIs?"

"I need to talk to someone about our Meta creative strategy — who should I contact?"

Attribution

"We're seeing a big discrepancy between Meta reported conversions and GA4. Can you explain why that might be based on our attribution setup?"

"Should we use last-click or data-driven attribution for our Google Shopping campaigns?"

Reporting

"Write a weekly performance report for the performance team for the week of May 26."

"Generate an executive summary of our Q2 paid media results."

Assets

"What video specs do I need for TikTok? What about YouTube?"

"Where do I find the Q2 brand campaign assets and what's the naming convention?"

"Are we ready to launch the Meta prospecting campaign? Do we have all the required creative sizes?"

Testing

"What have we learned from our Meta creative tests this year?"

"We want to test bid strategies on Google Ads — has anyone tested tROAS vs tCPA before? What happened?"

"What should we test next on LinkedIn? We haven't run any tests there in 6 months."

"Is the TikTok creator test statistically significant yet?"

Audiences

"What first-party audiences do we have available on Meta right now?"

"Explain our lookalike strategy — what seed audiences are we using and what expansion size performs best?"

"We have an Experian contract — are we using all the segments we're paying for?"

"What third-party audience layers should we apply to a new DV360 campaign targeting homeowners?"

Measurement

"Do we have server-side tracking on Meta? What's our Conversions API match rate?"

"What are our CM360 u-variables and what does u3 capture?"

"We're seeing a discrepancy between Meta CAPI and GA4 purchase counts — why might that happen based on our setup?"

"Is our data layer capturing order value correctly for all conversion events?"

"What measurement partners do we use and when is our next MMM run?"

Connecting to live APIs

The default FileAdapter reads from local JSON files. To connect to a live platform API:

Step 1 — Create a new adapter file, e.g. src/adapters/google-ads-adapter.ts

Step 2 — Implement the PaidMediaAdapter interface:

import type { PaidMediaAdapter } from "./base.js";
// ... import Google Ads API client

export class GoogleAdsAdapter implements PaidMediaAdapter {
  constructor(private customerId: string) {}

  async getCampaigns(filters = {}) {
    // call Google Ads API
  }

  // implement all other methods...
}

Step 3 — Swap it in src/index.ts:

// Before:
const adapter = new FileAdapter(DATA_DIR);

// After:
const adapter = new GoogleAdsAdapter({
  customerId: process.env.GOOGLE_ADS_CUSTOMER_ID!,
  developerToken: process.env.GOOGLE_ADS_DEVELOPER_TOKEN!,
});

All tools, resources, and prompts work unchanged — they only call the adapter interface.

Hybrid approach: You can also create a HybridAdapter that pulls campaigns from the API but falls back to file data for teams, attribution, and reporting templates (which rarely change and aren't in platform APIs).

Troubleshooting

Claude Desktop doesn't show the paid-media tools

  1. Check the config path: ~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows)

  2. Make sure paths in the config are absolute, not relative

  3. Check that npm run build ran without errors (dist/index.js must exist)

  4. Fully quit and reopen Claude Desktop (Cmd+Q on Mac, not just closing the window)

  5. Check Claude Desktop's MCP logs: Settings → Developer → Logs

Server starts but returns no data

The server logs warnings (to stderr) for each data file it can't load. Run the server directly to see them:

node dist/index.js

You'll see lines like:

[FileAdapter] Warning: Data file not found: /path/to/data/campaigns.json

Data changes aren't reflected

The server loads data files at startup. After editing any file in data/, restart the MCP server:

  • In Claude Desktop: Settings → Developer → restart the paid-media server

  • Or fully restart Claude Desktop

JSON syntax errors

If a data file has a syntax error, the server silently skips it. Validate your JSON at jsonlint.com or run:

node -e "JSON.parse(require('fs').readFileSync('data/campaigns.json', 'utf8'))"

Performance data not aggregating correctly

Rate metrics (CTR, CPC, ROAS) are re-computed from raw totals during aggregation. If you include both raw counts and pre-calculated rates in your records, the server will sum both, which will double-count rates. Best practice: include only raw count metrics in records (impressions, clicks, spend, conversions, conversion_value) and let the server compute rates.

Project structure

paidmedia-mcp/
├── src/
│   ├── index.ts                  # MCP server entry point
│   ├── types.ts                  # TypeScript interfaces for all data models
│   ├── adapters/
│   │   ├── base.ts               # PaidMediaAdapter interface — implement to connect live APIs
│   │   └── file-adapter.ts       # Default: reads from data/*.json at startup
│   ├── tools/
│   │   ├── campaigns.ts          # list_campaigns, get_campaign, list_accounts, get_account
│   │   ├── teams.ts              # list_teams, get_team, get_team_for_account, list_team_members, get_team_member
│   │   ├── performance.ts        # get_campaign_performance, get_team_performance, get_benchmarks
│   │   ├── attribution.ts        # list_attribution_models, get_attribution_model, compare_attribution_models
│   │   ├── reporting.ts          # list_reporting_templates, get_reporting_template, build_performance_report
│   │   ├── assets.ts             # get_asset_library, list_asset_categories, get_asset_category, get_asset_specs
│   │   ├── testing.ts            # get_testing_methodology, list_tests, get_test, get_test_learnings
│   │   ├── audiences.ts          # get_audience_library_overview, list_first_party_audiences, list_data_providers,
│   │   │                         #   get_lookalike_strategy, list_third_party_audience_layers, get_onboarding_platforms
│   │   └── measurement.ts        # get_measurement_overview, get_tag_management, list_pixels_and_tags,
│   │                             #   list_conversion_apis, get_cm360_setup, get_website_data_capture, list_measurement_partners
│   ├── resources/
│   │   └── index.ts              # 9 readable MCP resources
│   └── prompts/
│       └── index.ts              # 9 pre-built prompt templates
├── data/
│   ├── metadata.json             # Company info
│   ├── accounts.json             # Ad accounts by platform
│   ├── teams.json                # Media teams, objectives, KPIs
│   ├── team-members.json         # Team members, roles, responsibilities
│   ├── campaigns.json            # All campaigns
│   ├── historical-performance.json  # Daily metrics + benchmarks
│   ├── attribution-models.json   # Attribution configurations
│   ├── reporting-templates.json  # Report structures by audience
│   ├── assets.json               # DAM info, asset categories, per-platform specs
│   ├── testing.json              # Testing methodology, tools, and test history
│   ├── audiences.json            # 1P audiences, data providers, LAL strategy, 3P layers
│   └── measurement.json          # TMS, pixels, Conversion APIs, CM360, data layer, partners
├── claude_desktop_config.example.json  # Config snippet for Claude Desktop
├── package.json
├── tsconfig.json
└── README.md

Contributing

Issues and pull requests welcome. If you build an adapter for a specific platform (Google Ads API, Meta Marketing API, etc.), consider contributing it back.

License

MIT

This server cannot be installed

F
license - not found
-
quality - not tested
C
maintenance

How are these scores calculated?

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/arcticgreyy/paidmedia-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server