Mindcase for Outbound Prospecting

The LinkedIn lead-gen walkthrough on this blog shows the pipeline by hand: find companies, pull the people at them, filter down to a shortlist, optionally check who's active. Every step in that post is a real API call you can paste into your own code.

This post is about skipping the paste-it-in-yourself part. If you're already working with a coding agent — Claude Code, Cursor, anything that reads a skills file — you can hand it the same pipeline and let it run the calls itself.

The full machine-readable schema for every endpoint below lives at mindcase.co/skills.md. An agent reads that file once and knows the inputs, outputs, and price of every Mindcase API without you explaining any of it.

What does it mean to hand prospecting to an agent?

Normally, turning "get me brand managers at D2C companies in New York" into working code means: read the docs, figure out the request shape, write the loop, handle pagination, then repeat that for three more endpoints. An agent that's read /skills.md already knows all four of those request shapes — company search, employee search, profile search, email lookup — so the step you'd normally spend writing code disappears. You describe the target, the agent picks the right endpoint, runs it, and hands you rows.

/skills.md isn't a marketing page the agent skims — it's a literal agent index: platform-plus-task names (linkedin/companies, linkedin/profiles, linkedin/emails) each with their real input fields, output fields, and price. "Find companies, then people, then emails" maps directly onto that index without you naming a single endpoint — the agent reads the request, matches it to the closest agent block, and confirms the inputs it's missing before it spends any of your wallet.

How do you set up an agent to run this?

Point your agent at the skills file once, then give it the actual ask in plain language:

Set up mindcase.co/skills.md, then use Mindcase to find brand managers at
D2C companies in New York, get their work emails, and give me a shortlist
of who's worth messaging first.

That's the same "copy prompt" pattern used across this site — the agent reads the skills file, picks linkedin/companies to find the accounts, linkedin/profiles to find the people, and linkedin/emails to resolve work emails, without you naming any of those endpoints yourself.

What does the agent actually call, step by step?

Under the hood it's the same three calls from the tutorial post, plus one more. Here's the piece that post didn't cover — turning a shortlisted profile into a work email:

curl -X POST "https://api.mindcase.co/v1/data/linkedin/emails/run?wait=true" \
-H "Authorization: Bearer $MINDCASE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "params": {
    "urls": [
      "https://www.linkedin.com/in/jane-doe",
      "https://www.linkedin.com/in/john-smith"
    ]
  }
}'

# $0.02 per email returned

The agent runs this only on the rows that survived the shortlist step, not the whole raw list — LinkedIn Email API is priced per email, so filtering first is a cost decision as much as a quality one. Feed it a company or profile URL and it fails; it only resolves an actual LinkedIn profile URL to a work email, two fields back: profileUrl, email.

How do you keep a human in the loop?

The agent stops at the shortlist and the email list — it doesn't send anything. That's a deliberate line: Mindcase returns data, not outreach. Sending a message is still a decision you make, on your own tools, after you've looked at who's on the list. An agent that both finds your leads and emails them without a review step is one bad prompt away from a compliance problem.

This matters more the further the pipeline runs unattended. A one-off "find me 20 brand managers" prompt is easy to eyeball before you act on it. A recurring pull that refreshes your target list weekly is exactly the setup where a bad filter or a stale ICP compounds quietly — worth a final human skim before each batch goes anywhere near an outreach tool, not just the first time you set it up.

What can't the agent do for you?

Two cases.

A profile with no distinguishing field gets ranked low, not invented a reason to rank high. If a profile has no recent posts, no job-change flag, nothing beyond a title and a name, the agent's judgment only goes as far as the real fields returned — it can deprioritize a thin profile, but it can't manufacture signal that was never in the data.

Not every profile resolves to an email. The endpoint returns rows only for the profiles it actually finds a work email for, so an agent running this against 200 shortlisted profiles should expect fewer than 200 emails back — a partial result here is the normal outcome, not a bug to retry.

Which endpoints does the agent call, and what do they cost?

FAQ