MCP server

The GetProspect MCP server lets an AI assistant search people and companies, enrich contacts and verify emails in your workspace, using the same data and credits as the API.

• Server URL: https://api.getprospect.com/mcp
• Transport: Streamable HTTP
• Authentication: sign in with GetProspect (OAuth), or an API key in the x-api-key header

The tools call the same v2 endpoints as the REST API, so rate limits and credits follow the same policy.

Authentication

Sign in with GetProspect is the default. The client discovers the server's OAuth metadata and registers itself (OAuth 2.1 with PKCE and dynamic client registration), then opens a GetProspect page where you sign in and pick the workspace it may use. You paste nothing. The assistant acts as you, in that workspace, and spends that workspace's credits.

Access tokens last 15 minutes and refresh on their own. A connection that is used at least every 60 days stays signed in.

API key is for clients that only take headers. Create a key on the API page and send it as x-api-key, as for the REST API (see Authentication). The key acts in the workspace where you created it.

Credentials go in headers only. The server refuses an API key in the query string, and an API key must be sent as x-api-key, not as Authorization: Bearer (that header carries OAuth tokens).

A missing, expired or revoked credential returns HTTP 401, which makes a signing-in client start the sign-in again.

Add it to your assistant

The same steps are on the Integrations page of the app, with copy buttons. Menu names change as clients update; the URL is always the one above.

Claude (web and Desktop)

1. Open Settings, then Connectors, then Add custom connector.
2. Paste https://api.getprospect.com/mcp.
3. Click Connect and sign in with GetProspect.

Claude Code

claude mcp add --transport http getprospect https://api.getprospect.com/mcp

Then type /mcp in Claude Code and sign in.

ChatGPT

1. Open Settings, then Apps & Connectors, then Advanced, and turn on Developer mode.
2. Create a connector, paste the server URL and choose OAuth.

Developer mode needs a paid ChatGPT plan.

Cursor

Use the Add to Cursor button on the Integrations page, then sign in when Cursor asks. Or add it to mcp.json with an API key:

{
  "mcpServers": {
    "getprospect": {
      "url": "https://api.getprospect.com/mcp",
      "headers": { "x-api-key": "YOUR_API_KEY" }
    }
  }
}

VS Code

Use the Add to VS Code button on the Integrations page, then sign in when VS Code asks. Or add it to mcp.json with an API key:

{
  "servers": {
    "getprospect": {
      "type": "http",
      "url": "https://api.getprospect.com/mcp",
      "headers": { "x-api-key": "YOUR_API_KEY" }
    }
  }
}

Grok

Open grok.com/connectors, choose New connector, then Custom, paste the server URL and sign in. A paid Grok plan may be required.

Perplexity

Open Settings, then Connectors, then Add connector, then Custom (remote). Paste the server URL and choose OAuth and Streamable HTTP. A paid Perplexity plan may be required.

Other clients

Any MCP client that takes a server URL can sign in the same way. If yours only takes headers, use the mcpServers shape shown for Cursor.

Tools

The server offers 16 tools. Free tools never spend credits.

Not found is free. An email your workspace already owns is free again anywhere. See Credits and rate limits for the full rules.

Results tables

In clients that support MCP Apps (Claude, ChatGPT and VS Code), search_people, search_companies, enrich_people, enrich_companies, save_people and save_companies show their results as an interactive table in the GetProspect style: company logos, email status, Reveal email, row selection and paging. Selecting rows and choosing an action sends a message to your assistant, so a paid action still shows its credit quote first. Paging a search is free. Other clients get the same results as text.

How credits work

Searching, filter lookups, lists and company lookups are free. Only the paid tools in the table spend credits.

Quote before spend. Before a paid tool runs, the server works out its worst-case cost. Each person costs at most 1 email credit and 1 phone credit, and every item not known to be owned counts.

• At or under 10 credits per call (all credit types together), the call runs straight away.
• Above 10 credits, the call spends nothing and returns a quote with the worst case and your balance. The assistant should show it to you and ask. Once you agree, it calls the tool again with max_credits set to at least the quoted amount.
• A call whose worst case is above max_credits is refused and spends nothing. Because the quote is an upper bound, max_credits at or above it caps what this call can charge. A bulk job charges as it runs and stops early if the balance runs out.

The 10-credit threshold is the default. estimate_cost shows the current quote.

Actual spend is usually lower than the quote: emails you already own and lookups that find nothing are free. Paid results report credits_used and your balance when it can be read. For a bulk job, get_bulk_enrichment reports the credits used.

estimate_cost returns the same quote on request, so the assistant can check before it asks you. It takes the paid tool's name (action) and the arguments that tool would get.

Quotes are on trust

The server cannot prove that you saw a quote. The protection is the quote contract in the tool descriptions, the approval prompt most clients show for tools that are not read-only, and the max_credits cap. Review a tool call before you approve it.

For more than 50 people, the assistant should use start_bulk_enrichment instead of repeated enrich_people calls. Poll the job with get_bulk_enrichment.

Prompts

The server also offers three ready-made prompts. Clients that support MCP prompts show them in a menu.

Usage and limits

• MCP requests are their own source. They show as MCP on the Usage page, in the source filter and the breakdown.
• MCP calls follow the same rate-limit policy as the REST API (see Credits and rate limits). Moving to MCP does not raise them. A limited call returns a tool error with the seconds to wait.
• Results are capped at about 20,000 tokens per call. When a result is cut, the text says what was cut and how to narrow the search.

Disconnect

Open Settings, then Security, and find the assistant under Active sessions, labeled with its name and "(connected app)". Click Disconnect. The assistant can no longer refresh its sign-in; an access token already issued can keep working for up to 15 minutes. An API key stops working when you delete it on the API page.

A user can have up to 10 connections (sign-ins) per workspace, shared by MCP assistants, Zapier and the Sheets add-on. Connecting another signs out the oldest. Browser sign-ins are not affected.

Troubleshooting