Digital Marketing Skills — MCP Platform

A Model Context Protocol (MCP) server and a curated library of digital-marketing skills. It lets AI assistants discover and apply industry-standard marketing methodology — strategy, research, SEO, paid media, content, email/CRM, CRO, analytics, reporting, budgeting and industry playbooks — through structured resources, prompts and tools, with safety and human-in-the-loop controls built in by default.

Marketing outcomes are never guaranteed. This platform encodes method, transparency and safeguards — not promises. See DISCLAIMER.md.

What's inside

• 180 skills across 31 categories (skills/00-foundation … skills/30-optimisation-disciplines), each a single rich SKILL.md with YAML metadata, methodology, decision rules, quality & compliance checklists, failure conditions, worked examples, KPIs and MCP mappings.
• MCP server (TypeScript/Node) exposing:
  • Resources — skill & workflow catalogues, individual skills/workflows, metrics/channels/ funnel/industry catalogues, per-industry playbooks, and advertising-policy summaries (marketing://policies/catalogue, marketing://policies/{network}) (marketing://…).
  • Prompts — 30 reusable, structured marketing prompts (audit, strategy, GTM, SEO cluster, Google/Meta ads, email nurture, reporting, budgeting, industry growth plans, compliance).
  • Tools — read-only discovery/metric tools plus controlled write tools that respect dry-run and approval policies.
• 8 multi-step workflows (workflows/) with explicit approval gates before spend/sends.
• Deterministic engines — pure, auditable metric calculations, a policy engine, an approval engine, an audit log and a workflow planner.
• Optional Python utilities (python/dm_skills) — metrics and experiment statistics (sample size, two-proportion z-test), dependency-light and reproducible.
• Governance & safety — Apache-2.0 licence (code) with a separate name/logo trademark notice, security model, compliance-first defaults, schemas, CI, and a deterministic evaluation harness. See docs/repository-protection.md for branch protection and ownership guidance.

Safety model (defaults)

• Read-only by default; writes are dry-run by default.
• Human approval required for high-risk, cost-incurring or externally side-effecting actions.
• Privacy by design — data minimisation, consent checks, PII redaction in logs.
• No fabricated data and no guaranteed results — assumptions and ranges are explicit.
• Mock connectors ship by default; live connectors are opt-in via config.

See SECURITY.md, docs/approval-model.md and config/.

Quick start

Requirements: Node.js ≥ 20.

npm ci
npm run build

# Run over stdio (for MCP clients)
npm run start:stdio

# Or run the Streamable HTTP transport
npm run start:http   # health check at http://127.0.0.1:8080/health

Copy .env.example to .env to adjust transport, safety defaults and directories.

Install as a Cursor plugin

This repository is a Cursor plugin: .cursor-plugin/plugin.json is the manifest and mcp.json registers the MCP server (launched via npx, so the bundled skills are resolved automatically — no working-directory setup needed).

• Local (private) install: copy or symlink this repo into ~/.cursor/plugins/local/, then run Developer: Reload Window.

  ln -s "$(pwd)" ~/.cursor/plugins/local/digital-marketing-skills

• Marketplace: publish the public repo at cursor.com/marketplace/publish (manual review).

Use across MCP clients

Once the package is published to npm (npm publish), any MCP client can run it via npx with no local checkout. The server bundles its skills/config and resolves them relative to itself.

Cursor / Claude Desktop / Gemini CLI (stdio):

{
  "mcpServers": {
    "digital-marketing-skills": {
      "command": "npx",
      "args": ["-y", "digital-marketing-skills-mcp@latest"]
    }
  }
}

• Claude Desktop: add the block above to claude_desktop_config.json.
• Gemini CLI: add it under mcpServers in ~/.gemini/settings.json.
• ChatGPT (and other remote-only clients) require a hosted HTTPS endpoint rather than a local command — see the next section.

Hosted HTTPS endpoint (remote clients)

Deploy the Streamable HTTP transport behind HTTPS (see docs/deployment.md for Vercel / Render / Fly / Docker). The endpoint is /api/mcp on Vercel, or /mcp when self-hosted. When DM_MCP_HTTP_AUTH_TOKEN is set, the token may be supplied either as an Authorization: Bearer <token> header or as a ?key=<token> query parameter.

Cursor / Claude (web or desktop) / VS Code — header auth:

{
  "mcpServers": {
    "digital-marketing-skills": {
      "url": "https://<project>.vercel.app/api/mcp",
      "headers": { "Authorization": "Bearer <DM_MCP_HTTP_AUTH_TOKEN>" }
    }
  }
}

ChatGPT (developer-mode app) — query-parameter auth:

ChatGPT's app form only offers OAuth and No authentication (no static-header field), so pass the token in the URL and select No authentication:

1. Settings → Security and login → enable Developer mode.
2. Settings → Plugins (or chatgpt.com/plugins) → + / Create:
   • Name: Digital Marketing Skills
   • Server URL: https://<project>.vercel.app/api/mcp?key=<DM_MCP_HTTP_AUTH_TOKEN>
   • Authentication: No authentication
   • Tick the risk acknowledgement, then Create.
3. In a chat, click + → Developer mode → enable the app (required per conversation).

From source (no npm)

{
  "mcpServers": {
    "digital-marketing-skills": {
      "command": "node",
      "args": ["/absolute/path/to/dist/server/index.js", "--transport", "stdio"]
    }
  }
}

Repository layout

skills/            # 180 SKILL.md files across 31 categories
workflows/         # multi-step, approval-gated workflow definitions (YAML)
industries/        # per-industry playbooks (saas, ecommerce, real-estate, retail)
templates/         # reusable output templates (plans, briefs, reports, approvals)
examples/          # end-to-end worked examples (illustrative)
config/            # metrics, channels, funnel, risk, approval, regional catalogues
schemas/           # JSON Schemas for skills, workflows, campaigns, reports, etc.
src/               # MCP server (core engines, connectors, mcp resources/prompts/tools)
python/            # optional dm_skills utilities + tests
scripts/           # validate-skills, evaluate
evals/             # evaluation harness docs + behavioural golden cases
docs/              # MkDocs documentation

Development

npm run typecheck        # TypeScript type-check
npm run lint             # ESLint
npm test                 # Vitest unit tests
npm run validate:skills  # validate skill metadata & structure
npm run evaluate         # deterministic structural evaluation

# Python (optional)
pip install -e ".[dev]"
pytest && ruff check python && mypy python/dm_skills

Skill catalogue (categories)

00-foundation, 01-market-research, 02-customer-research, 03-brand-positioning, 04-marketing-strategy, 05-offer-and-messaging, 06-content-marketing, 07-seo, 08-paid-search, 09-paid-social, 10-organic-social, 11-email-marketing, 12-crm-lifecycle, 13-conversion-optimisation, 14-experimentation, 15-analytics-measurement, 16-reporting-insights, 17-budgeting-forecasting, 18-marketing-automation, 19-compliance-privacy, 20-saas-growth, 21-product-led-growth, 22-ecommerce-marketing, 23-real-estate-marketing, 24-retail-marketing, 25-account-based-marketing, 26-community-reputation, 27-content-production, 28-platform-playbooks, 29-advertising-policy, 30-optimisation-disciplines.

Contributing

See CONTRIBUTING.md, GOVERNANCE.md and CODE_OF_CONDUCT.md. New skills must pass npm run validate:skills and npm run evaluate.

Licence

Apache-2.0.