Playbooks · Skill & signal
Connect Claude to the hosted MCP server, describe the list you want, and let it search, find and verify while you watch each call. Then you export a CSV and upload it to Smartlead yourself. We do not sync sequences and we do not send mail.
Last updated September 22, 2026
Pattern: Claude + MCP → search, find, verify in conversation → CSV → your Smartlead upload
Credit math: Worked example at stated counts: a conversational run over 40 people. people_search returning 40 results at 0.1 = 4 credits. 40 email_finder calls at 5 = up to 200 credits, less every miss, which bills zero. 26 email_verifier calls at 2 = 52 credits. An agent that re-runs a step because you rephrased the request pays for it again. 0.1 credits per people_search result, 5 credits per email_finder call, 2 credits per email_verifier call
ROI math: This is a build-the-list workflow, not a scale workflow. It is worth it when the list is small enough that a conversation beats writing a script, and it stops being worth it the moment you would rather loop.
Owned internally by: A founder or an AE doing their own prospecting, with no script in between
Best for: Small, high-judgement lists where you want to look at each account before it goes into a sequence
// ~/.claude.json — or the equivalent MCP config in Cursor, Windsurf, VS Code
{
"mcpServers": {
"richapi": {
"url": "https://mcp.richapi.ai/mcp",
"headers": { "x-api-key": "YOUR_KEY" }
}
}
}
Find heads of demand generation at Series B B2B SaaS companies in the UK.
Pull about forty people, get their work emails, and verify every one you find.
Drop anyone whose address fails verification or comes back as a catch-all.
Then give me a CSV with first name, last name, company, title and email,
with the column headers Smartlead expects.
Sign up at app.richapi.ai and copy your key.
25 free credits, no card.
Add the hosted MCP server at mcp.richapi.ai to Claude, Cursor, Windsurf or VS Code.
Nothing to install.
Ask for the list in plain language and approve the tool calls as they come.
Watch the first few rather than approving blind.
Tell it to verify before it hands you anything, and to drop failures — otherwise you will paste unverified rows into a sequence.
Ask for a CSV with Smartlead's column headers, then upload it to Smartlead yourself and write the sequence there.
There is no Smartlead integration. We do not sync sequences, we do not write campaigns, we do not send mail, and we have no plans on this page to suggest otherwise. [Sending and sequencing are not things we do](/platform) — we are the data layer. So the seam in this workflow is a file. Claude builds the list, you download a CSV, you upload it in Smartlead, you write the sequence there. That is a manual step and it takes about a minute. Most sequences underperform because the list went in unverified, not because of the sending tool. The part of this workflow that is automated — search, find, verify, drop the failures — is the part that decides whether the campaign works. The part that is manual is the part that takes a minute. **[Get 25 free credits — no card](https://app.richapi.ai)**
The MCP server is hosted at `mcp.richapi.ai`. There is nothing to install and no local process to babysit — add the server, paste your key, and the RichAPI endpoints appear as tools in Claude, Cursor, Windsurf or VS Code. ```json { "mcpServers": { "richapi": { "url": "https://mcp.richapi.ai/mcp", "headers": { "x-api-key": "YOUR_KEY" } } } } ``` Then describe the list. The model picks the tools, and you approve each call. That approval step is not friction, it is the cost control — every tool call is a credit decision and you get to see it before it happens.
The instruction that matters most in the prompt is the one about verification. Left to itself, an assistant will happily return a tidy table of addresses it found and never checked, and a tidy table is exactly what makes you paste it straight into a sequence. Say it explicitly: verify every address, drop anything that fails, and tell me which ones came back as catch-all rather than silently including them. What to do with those is a separate decision — see [catch-all email handling](/use-cases/catch-all-email-handling) — but it should be your decision, not a side effect of a model's formatting preferences. `email_finder` and `email_verifier` are both multi-provider waterfall endpoints, so the lookups that fail come back with `billed: false` and cost zero. The searching step is not: `people_search` bills per result returned.
| Step | Count | Rate | Credits | |---|---|---|---| | `people_search` | 40 results | 0.1 / result | 4 | | `email_finder` | 40 calls, 26 hits | 5 / call, misses free | 130 | | `email_verifier` | 26 calls | 2 / call | 52 | Rates used above: 0.1 credits per `people_search` result, 5 credits per `email_finder` call, 2 credits per `email_verifier` call. Add the conversational overhead honestly: if you rephrase the request and the model re-runs the search, you pay for the second search too. Convert credits at your tier rate on [the pricing page](/pricing).
**The model re-runs expensive steps.** Ask for "the same list but only UK" and a model may redo the whole search instead of filtering what it has. Tell it to filter the results it already has, in the same message. **It formats what it could not find.** Assistants fill tables. If a row has no verified email, you want it dropped or flagged, not tidied into place. Ask for the count of rows dropped as part of the output — a run with zero drops is a run you should look at closely. **Your CSV headers do not match.** Smartlead expects its own columns. Ask for them by name in the prompt so you are not hand-editing a header row at upload time. **Expecting a sync.** There is no webhook from us into Smartlead, no campaign push, no status write-back. If you want the list to land in Smartlead without a human, that is an automation you build against Smartlead's own API — and at that volume you have outgrown a conversational workflow anyway. **Approving tool calls blind.** The approval prompt is the only thing between a vague request and a wide, per-result search. Read the first few.
For the scripted version of the same chain, see [list build and verify](/use-cases/list-build-and-verify), or browse [the use-case index](/use-cases). To let the assistant dig into one account rather than build a list, [agent-driven account research](/use-cases/agent-driven-account-research) is the better fit. Setup details for the MCP server live on [the Claude integration page](/integrations/claude), and [the email finder API](/api/email-finder) documents the billing behaviour the assistant is relying on.
**Does this push leads into Smartlead automatically?** No. You export a CSV and upload it. We have no Smartlead integration and do not sync sequences. **Do you send the emails?** No. We have no sending or sequencing product at all. Smartlead sends; we supply the rows. **Is the MCP server a separate product or separate billing?** Neither. Same key, same credits, same endpoints — a different way to call them. **How big a list is this good for?** Small enough to read. Forty rows is comfortable, a few hundred is tedious, a few thousand means you should be writing a script against the REST API instead. **What stops the assistant spending my credits?** Tool-call approval in your client, and the fact that you can see every call. The waterfall endpoints also return `billed: false` on a miss, so the lookups that fail are free — but the per-result searches are not, and a re-run search bills again. **Do I need a credit card?** No. 25 free credits on signup.
Connect the MCP server and ask for five people. You will know inside two minutes whether the conversational version suits you. **[Get 25 free credits](https://app.richapi.ai)**