Skip to main content
The MCP server is a thin, honest wrapper over the Generect API. Two rules make it safe to hand to an autonomous agent:
  1. Sizing an audience is free. You can learn how many people match an ICP, and what pulling them would cost, without spending anything.
  2. Every response reports what it actually cost. Each tool returns a cost block containing the API’s own meta.amount_charged, so an agent telling a human what it spent is quoting the biller, not guessing.
This page describes MCP server 0.13.0: interactive results arrived in 0.12.0, the five webhook tools replaced manage_webhooks in 0.13.0, and resources, prompts and the vocabulary guard date from 0.9.0. Remote users (https://mcp.generect.com/mcp) are always on the deployed version. Local users should run npx generect-ultimate-mcp@latest.

Free tools

Billable tools

Exact per-operation prices for your account are in billing settings, and get_balance returns them too. See Pricing for the list rates.

Filter values are matched exactly, and two of them are not validated

This is the most expensive thing to get wrong, so it is worth stating plainly. The search endpoints do not validate all filters the same way: company_industries: ["Fintech"] is not a LinkedIn industry, so it matches nothing and you get a perfectly successful count of zero. That is indistinguishable from a real empty audience, and it is how an agent ends up telling a user “there is nobody like that” when the truth is “that is not the name of an industry”. From 0.9.0 the MCP server checks these values before it sends anything:
  • an unknown industry, headcount bucket or company type is refused locally, with the closest valid names: Fintech → Financial Services, 50-200 → 51-200. Nothing is sent and nothing is charged;
  • a value that is merely mis-spelled or mis-cased is corrected (software development → Software Development), because matching is exact;
  • an unknown seniority or job function is a warning rather than a refusal: the engine does match loosely (Owner finds people even though the canonical label is Owner / Partner);
  • allow_unlisted_values: true overrides the check, for the case where the API has gained a value the server’s snapshot does not have yet.

Resources

The vocabularies are exposed as MCP resources, so a client can read the exact list once instead of guessing: Reading a resource is free and touches no data endpoint.

Prompts

The server also ships workflow prompts, which appear as slash commands in clients that support them: size_an_audience, build_prospect_list, enrich_my_list, spend_report. Each starts from the free step.

Spend ceiling

A row cap bounds how many results you get, not how much you pay: the per-row price depends on the mode and your tier. So any call whose worst case exceeds the server’s per-call ceiling (default $5, MCP_MAX_SPEND_PER_CALL) is refused with the exact figure, and has to be repeated with confirm_spend_usd set to at least that amount. This is checked before start_bulk_job submits, because a bulk job reserves its whole cost at submit time and cannot be called back.

Database or realtime

Every search and enrich exists in two modes: cached database (sub-second, cheaper, free counts, core filters) and live realtime (5–60s, pricier, billable counts, every filter). See Database vs Real-time. Tools take a mode parameter: Counting is deliberately stricter: a realtime count costs money, so count_leads and count_companies never run one implicitly. If your filters need the live index, the tool returns the blocking filter names and the price, and waits for you to ask again with mode: "realtime".

The order to call things in

Records carry a stable id across preview, search and enrich, so once a lead is in your list you never have to search for them again: pass the id to enrich_lead, generate_email or find_phone directly.
For repeatable or scheduled work, submit up to 50 records with start_bulk_job and register a webhook with create_webhook instead of polling. A submitted job runs to completion even if your balance later hits zero, because its cost was reserved at submit time. Only submit lists that have been approved.

Interactive results in Claude

In Claude (web, desktop and mobile), ChatGPT and other clients that support MCP Apps, results come back as cards in the conversation instead of raw JSON. Clients without it get the same text and structured results as before. Money inside the cards follows the same rules as the tools:
  • Nothing paid happens on the first click. “Find emails” first reads your real email price, shows the maximum for the people you selected, and spends only after you confirm. A miss is free.
  • Found emails go back to Claude, so “write to them” works without asking again. With a test key they are marked as fictional.
  • Claude asks before tools that can charge. Only tools that are free by default are marked read-only, which lets Claude run them without asking: counting, balance, health, bulk-job status and listing webhooks. The one paid case among them, a realtime count ($0.04 at the standard price), runs only when mode: "realtime" is requested explicitly. Every other tool needs your approval (“Always allow” works per tool, except for the three webhook tools below).
  • Webhook changes are confirmed every time. create_webhook, update_webhook and delete_webhook are marked destructive, so Claude asks before each call. A bulk-job completion event carries the job’s results (emails, phones, records), so these tools decide where your paid data is sent.

When to use the REST API instead

MCP is the right surface when a model is choosing what to fetch. Call the REST API directly when you need something MCP deliberately does not do:
  • pulling more than 100 rows in one request
  • pipelines with no model in the loop (nightly syncs, CRM backfills)
  • your own retry, concurrency and storage policy