Gifting API and MCP server for developers and AI agents

Built for software that retries

The API assumes agents and workflows will retry, run twice and act on stale data. Adds are idempotent, every error says what to do next, and anything that spends money is reserved for a signed-in owner or admin, or for limits they set.

• Idempotent adds. Send sourceSystem and externalId; a repeat returns alreadyQueued: true and never charges twice. create_campaign takes an idempotencyKey.
• Errors with a next step. Every error carries nextAction (open_url, reread_and_retry, ask_user, retry_later or stop) and retryable, sometimes with retryAfterMs.
• Spending stays with people. Keys and connections read, draft and add people. Approving, paying, adding funds and auto-approve are signed-in only.
• Clear limits. 240 requests a minute per workspace, 500 people per batch campaign and 2,000 per playbook.
• Signed webhooks. HMAC-SHA256 signatures, at-least-once delivery and up to eight attempts per event.

The playbook loop

1. search_gifts or GET /api/v1/catalog to choose a gift.
2. create_campaign with mode: "always_on" saves a playbook draft.
3. A teammate turns it on at the review link. Agents can’t.
4. add_to_playbook (POST /api/v1/playbooks/CAMPAIGN_ID/people) once per trigger. Read approval.state and approval.reason.
5. list_queue and get_balance show who waits and whether funds are low; hand the links to an owner or admin.
6. get_gift_link_records once a gift exists, and list_tracking_events or a signed webhook for status.

curl -X POST https://gtmswag.com/api/v1/playbooks/CAMPAIGN_ID/people \
  -H "Authorization: Bearer $GTM_SWAG_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "recipient": {
      "name": "Jamie Chen",
      "company": "Northstar",
      "title": "VP of Sales",
      "sourceSystem": "hubspot",
      "externalId": "12345",
      "addressMode": "known",
      "address": "100 Congress Ave",
      "city": "Austin",
      "state": "TX",
      "zip": "78701",
      "country": "US"
    }
  }'

A 200 response includes queued, recipientId, alreadyQueued, approvalUrl and approval: {state, reason}. Without auto-approve, the person waits with reason auto_approve_off and nothing is charged.

Reading the catalog and prices

The public catalog at /api/v1/catalog also lists retired gifts for history, so filter in this order: retired: false, then available: true with no priceQuoteRequired, and skip any gift with priceCents: 0.

• Price per gift is priceCents + shippingCents in USD cents, with applicable sales tax included.
• shippingQuoteRequired: true means delivery isn’t priced yet. Budgets don’t include it and auto-approve never approves the gift.
• requiresPortrait, requiresArtwork and personalizationText tell you what to ask the user for. Never invent them.

Documentation

• Agent gifting guide: the playbook loop, roles, prices, errors, gift links and tracking.
• API and MCP reference: every route and tool with inputs, results and examples.
• OpenAPI 3.1 schema and the public catalog endpoint.
• Connection guide: Claude, Claude Code, other MCP clients and REST keys.
• llms.txt: a short index for AI agents.