MCP server

Connect Claude Code, Cursor, ChatGPT, or any MCP-speaking client to your PAD workspace and operate it from your editor. Read counts, run agents, approve insights, ship campaigns — all without leaving the keyboard.

One-line install

Claude Code (CLI)

claude mcp add pad-ai \
  --transport http \
  --url https://pad.thevisualpay.com/api/mcp \
  --header "Authorization: Bearer pad_sk_YOUR_KEY"

That's it. Your next claude session has every PAD tool available. Generate a server-scoped API key at /dashboard/settings/api-keys if you don't have one.

Cursor / Windsurf / VSCode (config file)

Add this to .cursor/mcp.json(or your tool's equivalent):

{
  "mcpServers": {
    "pad-ai": {
      "transport": "http",
      "url": "https://pad.thevisualpay.com/api/mcp",
      "headers": {
        "Authorization": "Bearer pad_sk_YOUR_KEY"
      }
    }
  }
}

ChatGPT (Custom GPT action)

OpenAI's ChatGPT MCP support is still limited as of mid-2026 — for now wrap the MCP endpoint as a Custom GPT action against https://pad.thevisualpay.com/api/mcp using Bearer auth. Tool discovery happens via the OpenAPI we serve at /v1/openapi.json.

Auth

Same Bearer token as the REST API. Use a server-scopedkey — browser keys can't mutate. The key already binds to one workspace; the optional X-Pad-Workspace header only matters if you need to pin a specific slug when one key spans multiple workspaces.

Need to discover your slug from the API? Hit GET /v1/workspacewith the same Bearer token — it returns the workspace + organization + key scope. Useful in scripts that don't want to hardcode the slug.

Every mutation gets an audit-log entry tagged { type: 'system', source: 'mcp' }. You can audit the agent at /dashboard/settings/audit.

Tools

50 tools total covering every non-credential operation in the dashboard. Schema is exposed via the MCP tools/list JSON-RPC method; your client auto-discovers. Groups below are for human navigation only.

Example prompts

Drop these straight into Claude Code, Cursor agent, or ChatGPT once the server is connected. The agent figures out the right tool sequence.

• Daily standup, one prompt

  "Use workspace_summary, then list the top 3 pending insights I should look at. For each, suggest approve or dismiss with a one-line rationale."

• Triage a deliverability spike

  "Run workspace_health. If bounce rate is over 2%, pause every active campaign and tell me which addresses are bouncing the most. Don't auto-resume."

• Smoke-test a new event type

  "Send a track_event of type 'checkout.completed' for externalId 'qa-1' with properties {amount: 99}. Then count_events for that type in the last hour."

• Ship today's growth experiments

  "Run run_agent(growth). When it returns the insight id, read it back with recent_insights and either approve_insight or dismiss_insight with a reason."

• Reactivation campaign sanity check

  "List my segments. Pick the one that looks like dormant paid users. segment_size it. If under 500 profiles, pause the winback campaign — we don't have enough cohort."

• Author a new template from scratch

  "Use ai_draft_template to write a welcome email for a B2B SaaS that just signed up — friendly, one CTA pointing at /dashboard. Then preview_template with firstName='Cris' to verify it renders, and save_template as a draft with slug welcome-v2."

• Build a segment + campaign from a sentence

  "translate_segment('paid users who haven\'t logged in for 14 days'). If count is over 50, save_segment as dormant-paid. Then list_segment_profiles(dormant-paid, 5) so I can sanity-check the cohort."

• Ship a 3-step winback campaign

  "create_campaign('winback-3step', 'Winback 3-step', 'subscription.cancelled', { trigger: { eventType: 'subscription.cancelled' }, steps: [{kind:'send', templateSlug:'winback-1', variables:{}}, {kind:'delay', ms: 259200000}, {kind:'send', templateSlug:'winback-2', variables:{}}] }, 'draft'). Then list_campaign_runs once it's active to verify enrollments are happening."

• Debug a leaking funnel

  "campaign_analytics(welcome-drip). Tell me which step has the worst open rate. If a step's click rate is below 5% AND it has at least 100 sends, suggest a copy revision via ai_draft_template."

Rate limits + safety

• Per-key rate limits match the REST API. MCP tools that mutate go through the same paths (event ingestion, audit logging) as the SDK — same gates apply.
• MCP cannot create API keys, rotate Stripe credentials, change billing, or delete workspaces. Those are dashboard-only.
• The audit log captures every mutation with source: mcp. You can grep /dashboard/settings/audit to see what your agent did.

Endpoint metadata

GET https://pad.thevisualpay.com/api/mcp returns the server version + tool count in plain JSON — useful for health-checks and as a sanity test before wiring auth.