Connect RocketVerifier to Claude or ChatGPT: One URL, No API Key

You are in the middle of a conversation with an AI assistant about a list of 300 conference leads. You want to know which addresses are real before anything goes to a sequencer. Until recently that meant leaving the chat, uploading a file somewhere else, and pasting results back in.

The RocketVerifier MCP server lets the assistant do it directly. MCP, the Model Context Protocol, is the open standard assistants use to call external tools. The hosted RocketVerifier server now supports OAuth, which means connecting it takes one URL and one click on a consent screen. There is no API key to create, copy or store in the assistant’s settings.

What you need

• A RocketVerifier account. A new account comes with 100 free credits when you sign up with a work email, no card required.
• An assistant that supports remote MCP servers with OAuth. Claude and ChatGPT both do, under their custom connector settings.
• The server URL: https://api.rocketverifier.com/mcp

Connect it to Claude

Anthropic’s custom connector guide says custom connectors are available on every Claude plan, with Free users limited to one. The menu names below match that guide at the time of writing; Anthropic changes them from time to time.

Then ask something concrete: “Verify jane@acme.com and tell me why if it is not deliverable,” or attach a CSV and ask for a bulk job.

Connect it to ChatGPT

OpenAI documents custom MCP servers in its Apps SDK guide. Workspace policies apply, so on a business plan an admin may need to allow custom servers first.

Menu labels have moved several times as OpenAI reworked connectors into apps and plugins. If yours look different, search ChatGPT’s settings for “MCP” or “custom”; the URL and the OAuth choice are the parts that matter.

What happens behind the one click

You paste a URL and approve once. In between, the assistant performs a short discovery sequence defined by the MCP authorization specification, built on three OAuth standards.

• AI client
• api.rocketverifier.com
• app.rocketverifier.com
• You

1. 1

   MCP endpoint answers 401 and names its metadata api.rocketverifier.com

   WWW-Authenticate: Bearer resource_metadata=".../.well-known/oauth-protected-resource"
2. 2

   Protected resource metadata names the issuer (RFC 9728) api.rocketverifier.com

   "authorization_servers": ["https://app.rocketverifier.com"]
3. 3

   Authorization server metadata lists the endpoints (RFC 8414) app.rocketverifier.com

   GET /.well-known/oauth-authorization-server
4. 4

   The client registers itself (RFC 7591) app.rocketverifier.com

   POST /api/oauth/register → 201 { "client_id": "rvapp_..." }
5. 5

   You sign in, pick an account and click Allow You

   /oauth/authorize ... &code_challenge_method=S256&resource=https://api.rocketverifier.com
6. 6

   The client calls tools with its token AI client

   POST /mcp Authorization: Bearer rv_at_...

Steps 1 to 4 and 6 are automatic. The only thing you do is step 5, once.

The first request to /mcp has no credentials, so the server answers 401 with a header pointing to its protected resource metadata (RFC 9728). That document names https://app.rocketverifier.com as the authorization server. The assistant reads the server’s own metadata (RFC 8414) to find the endpoints, then registers itself as a new public client (RFC 7591) and gets a client_id.

From there it is the standard authorization code flow with PKCE, plus a resource parameter that ties the token to https://api.rocketverifier.com. The token the assistant receives only works for the RocketVerifier API and only for the scopes you approved. If you build MCP clients yourself, the MCP OAuth docs show every request and response.

What the assistant can do once connected

The hosted server exposes the same tools as the RocketVerifier MCP server package. The ones you will use most:

• verify_email checks one address in real time and returns its status (deliverable, undeliverable, risky or unknown), a precise sub_status such as mailbox_not_found or catch_all, and a catch_all_verdict for accept-all domains. 1 credit.
• create_bulk_job, get_job_progress and get_job_results run a list of up to 100,000 addresses as one job and page through the results. 1 credit per address; duplicates in a list are not charged.
• preview_bulk_job estimates a list’s bounce rate and the credits needed before you commit. Free.
• check_credits, mx_lookup, check_disposable and check_domain_auth answer account and DNS questions. Free.

Assistants are good at the follow-up work: grouping risky results by reason, explaining what a catch_all verdict means for your send, or drafting a short note to a sales rep about which leads to drop. The verdicts themselves come from the same engine as the dashboard and API, with live SMTP checks and no email sent to the address.

Staying in control

Every assistant you connect appears in Settings, Connected apps in your RocketVerifier dashboard, with the account it uses and the scopes you approved. From there you can disconnect it, and its tokens stop working at once.

Two settings on the consent screen are worth using. Pick an organization if the work belongs to your team, so usage lands on the right credits. And set a monthly credit cap if you plan to hand the assistant large lists: once it reaches the cap, its verification calls are refused until you raise it or the month resets.

Other assistants and tools

Any client that implements MCP authorization should connect the same way with the same URL. For assistants without OAuth support, the AI agents hub has setup guides that use an API key, including Claude, ChatGPT and Cursor. If you want to understand what each verification result means before you act on it, read how email verification works.