Use Plyto from Claude, or any AI assistant

1Connect with OAuth, or create an API key

2Add the server to your client

claude mcp add --transport http plyto https://plyto.ai/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

{
  "mcpServers": {
    "plyto": {
      "url": "https://plyto.ai/api/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

npx mcp-remote https://plyto.ai/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

3Talk to your CRM

REST API

If your system speaks HTTP rather than MCP, the same API key works against /api/v1. Send it as a bearer token.

# List contacts
curl https://plyto.ai/api/v1/contacts?limit=50 \
  -H "Authorization: Bearer plyto_at_..."

# Create one (existing email updates instead of duplicating)
curl -X POST https://plyto.ai/api/v1/contacts \
  -H "Authorization: Bearer plyto_at_..." \
  -H "Content-Type: application/json" \
  -d '{"firstName":"Dana","email":"dana@example.com"}'

# Anything else: every tool is reachable by name
curl -X POST https://plyto.ai/api/v1/tools/crm.tasks.create \
  -H "Authorization: Bearer plyto_at_..." \
  -H "Content-Type: application/json" \
  -d '{"title":"Follow up with Dana"}'

• Resources: /contacts, /companies and /deals support GET and POST; a single contact supports GET, PATCH and DELETE.
• Everything else goes through POST /api/v1/tools/{name} with the tool's arguments as the JSON body.
• Paging: limit (max 200) and offset. Responses carry pagination.hasMore, measured rather than estimated.
• Errors are JSON with a readable message. Invalid arguments return 422 with the offending field named.
• The same rules as MCP apply: starting new spend and bulk sends are not reachable with a key. Single-record operations (including DELETE on a contact) follow the key's role, adjustments to campaigns a human already launched are allowed, and every call lands in the audit log.

Safety model

• Every call is scoped to one business and one user role; an external assistant can never see or touch another business.
• Starting new spend is blocked over MCP: there is no API path that launches a campaign, publishes a site, or bulk-sends email. Assistants draft; a human confirms those in the Plyto app. Ordinary record operations, including deleting a single contact or removing a tag, are available when the key's role allows them, and every one is written to the audit log.
• On campaigns a human has already launched, an assistant can adjust live settings: budgets, bidding, keywords, negative keywords, landing pages. It can also send an email to a single contact. Every one of these is written to the audit log and bounded by the spending caps and approval settings you set in the app, so grant keys accordingly.
• Every MCP tool call is written to the business audit log with the key that made it. Revoke any key instantly in Settings → API keys.

Protocol details

Streamable HTTP (JSON-RPC over POST) at https://plyto.ai/api/mcp. Protocol versions 2024-11-05 through 2025-06-18. Capabilities: tools. Auth: Authorization: Bearer plyto_at_… (thex-api-keyheader works too). The server is reachable from the public internet, including Anthropic's infrastructure for claude.ai connectors. Machine-readable metadata: server card.

Bring your own system of record

If your revenue lives outside Plyto (your own Stripe, HubSpot, a billing backend), you can still close the loop between ad clicks and money. Capture leads through your business's inbound webhook, then report what became of them to the same URL plus /outcome:

POST /api/inbound/{token}/outcome
Content-Type: application/json

{"email": "lead@example.com", "outcome": "customer"}

Plyto promotes the contact and reports the conversion back to the ad platform against the original click, so bidding optimizes toward people who pay rather than people who fill in forms. Outcomes are upgrade-only, so retries and replays are safe. Connected Stripe accounts do this automatically on the nightly sync; the webhook is for everything else. Your webhook URL lives in Settings → Integrations.

The same token also syncs contact data. POST to /properties with {email, properties: {plan: "growth", lifetime_revenue: 1240}} and Plyto writes every key that exists as a custom contact property, reports unknown keys back, and emits a property-changed event automations can trigger on.

Find Plyto in MCP directories

The same server is listed on Smithery, mcp.so, and Cursor Directory, plus the official MCP registry as ai.plyto/crm.