Panther Hook Blog
Integrations & Automation

Manage Your Lead Workflow with ChatGPT, Claude and Antigravity Using MCP

How remote MCP lets you inspect forms, diagnose webhook deliveries and manage authorized lead workflows in Panther Hook with ChatGPT, Claude or Google Antigravity.

AUTHENTICATED AI TOOLSChatGPTClaudeAntigravityOAuth / MCPWorkspaceSCOPED ACCESS

From dashboards to conversations, without surrendering control

Imagine a lead form stops reaching a customer-facing integration. You can open several dashboard tabs, inspect delivery attempts and compare failures manually. Or you can ask an AI assistant to list the failing deliveries in a particular Workspace and summarize what needs attention. With an authorized Panther Hook MCP connection, the assistant can use real application tools rather than inventing answers from a product description.

Panther Hook supports direct remote MCP connections with ChatGPT, Claude and Google Antigravity. Connecting does not depend on a marketplace plugin. You authorize access using the Panther Hook OAuth screen, select a Workspace and approve scopes. The tools available after connection depend on the client and the permission granted.

Connect your AI assistant to Panther Hook

What the Model Context Protocol actually changes

MCP is a standardized way for a compatible AI client to discover and call tools provided by another service. Instead of pasting an API key or uploading lead exports to an assistant, you connect an authenticated remote server. Panther Hook can then serve scoped tool operations for real Workspace data.

This is not the same as granting an AI unrestricted control of your account. OAuth identifies the user and grants, Panther Hook validates the current Workspace membership and role, and individual tools enforce their own data and plan restrictions. Tool discovery is not evidence that every tool call is allowed.

Three layers in a Panther Hook MCP session
LayerResponsibility
AI clientChatGPT, Claude or Antigravity chooses an appropriate exposed tool
OAuth connectionYou authorize specific permissions for a selected Workspace
Panther Hook backendEnforces role, plan, scope, quotas and required approvals

Three practical workflows worth trying

An AI assistant can accelerate a workflow by chaining several small, authorized reads and writes, but there are boundaries. The exact available tool set can change, data may be summarized for privacy, and a write can require a stronger role or an additional confirmation. Do not treat incomplete responses as complete exports or a pending approval as a successful change.

  • “Show the forms in my authorized Workspace.” Use get_user_info, list_workspaces and list_forms to confirm the identity and available resources before taking action.
  • “Which webhook deliveries are failing?” Have the assistant inspect permitted delivery statuses and retry information. Check the actual response rather than trusting a guessed diagnosis.
  • “Help me set up a form for a new campaign.” Describe the form and lead pipeline, review the plan and client/project context, then approve supported write operations only when you mean to create resources.

Direct connection to ChatGPT, Claude or Antigravity

All three use the same Panther Hook remote MCP endpoint. The client-specific setup differs: ChatGPT requires an account configuration that supports custom MCP connections, Claude provides custom connectors, and Google Antigravity has an MCP manager. Each then starts Panther Hook OAuth to bind a particular Workspace and requested scopes.

Once the connection is authorized, test a read call. If OAuth completes but no tools are available, inspect the client’s tool discovery rather than repeatedly creating new authorizations. Marketplace distribution is a separate convenience, not the transport needed for direct MCP use.

text
https://ph-mcp-gw-9qryt8yw.ew.gateway.dev/mcp
Step-by-step ChatGPT, Claude and Antigravity setup

Security: permissions are part of the feature

Ask only for the access needed to complete the task. A read-only workflow should not ask for write scopes. Sensitive actions use Panther Hook’s explicit approval mechanisms where required, and approval links open the authenticated first-party interface.

Customer lead contents, notes and outside webhook payloads may contain untrusted text. The assistant must treat those as data, not instructions. Keep authentication codes, API tokens, personal lead details and provider secrets out of prompts. Review the connected applications and revoke any authorization that is no longer necessary.

Understand Workspace roles and access

A productive starting point

Start with an existing Workspace and a read-only request. Confirm that the assistant knows which Workspace it is authorized to access, then inspect a harmless list of forms or delivery statuses. Expand to writes only after you are comfortable with how your chosen AI client presents permission prompts and Panther Hook approvals.

Panther Hook’s account settings include External AI access, where you can copy the remote MCP URL, inspect active connections and disconnect them. There is no reason to wait for a public plugin listing before using a supported native MCP connection.

Open MCP connection instructions