Solutions

Claude for Recruiting

Build your sourcing twin in Claude

Claude for Recruiting: Source, Score and Shortlist Candidates on Live Data

Claude for Recruiting: Source, Score and Shortlist Candidates on Live Data

Claude for Recruiting: Source, Score and Shortlist Candidates on Live Data

Claude has no candidate database of its own. Connect it to Crustdataโ€™s MCP server and it searches 1B+ people and 60M+ companies from 15+ sources, scores each match against your brief with a one-line reason, and learns from your feedback between batches. Crustdata retrieves data live when you ask, and its database refreshes every 1 to 30 days. You review every shortlist and make every call.

ellips

See Claude Source Candidates on Live Crustdata Data

See Claude Source Candidates on Live Crustdata Data

A real search with the Sourcing Skill: Claude searches live Crustdata profiles,

scores the matches, and learns from your feedback in the next batch.

69%

Reduction in sourcing time

A full day to 2.5 hrs per role

69%

Reduction in sourcing time

A full day to 2.5 hrs per role

8-10%

Candidate interest rate

vs 2.3% industry average

50%

Sourcing capacity increase

No new hire

50%

Sourcing capacity increase

No new hire

How to Use Claude for Recruiting, Step by Step

Each step runs in one Claude conversation once the Crustdata connector is on (setup is further down). No code is needed. Paste the prompt, swap in your role, and open the linked guide when you want the full method.

Step 1. Turn the Job Description Into a Rubric

Paste the job description and your intake notes, and ask Claude for rules it can score against: hard gates that drop anyone who fails, weighted signals that add up to 100, and a few exclusions. Add two or three people the hiring manager liked and say why. Read the rubric before any search runs, because every batch after this is scored against it.

Here is the job description and my notes from the intake call.
Turn them into a sourcing rubric with hard gates, weighted signals that add up to 100, and exclusions.
Ask me up to five questions about anything vague first.
Don't search yet.

Step 2. Size the Pool, Then Pull a Small First Batch

Crustdata returns a count of everyone who passes the gates, so you can fix the gates before you open a single profile. If the count runs into the thousands, tighten a gate. Then pull 25 people. Each one comes back with a score and a one-line reason, which shows you how Claude read the rubric.

Search Crustdata using only the hard gates and tell me how many people match.
If the pool is too big to review, suggest one gate to tighten.
Then pull the top 25 and score each person 1 to 4 with a one-line reason.

Step 3. Correct the Batch in One Message

When the wrong kind of person keeps showing up, the rubric has a gap. Describe the pattern across the whole batch in one message, change one or two rules per round, and stop when a round adds no new top scores.

Rows 2, 6 and 14 manage large teams, and this role is hands-on.
Rows 5 and 8 are right, because they shipped [product type] at companies under 200 people.
Turn that into rubric rules and show me what changed.
Then pull the next 25 and skip anyone you've already shown me.

Step 4. Test the Shortlist Against People You Know

Give Claude a few people who clearly qualify, such as past placements. If several are missing, a rule is too strict, and Claude can tell you which one. If the pool runs dry, search by target companies or past employers instead.

Here are 8 people who should qualify for this role: [names and companies].
Tell me which ones are in the results.
For each one that isn't, tell me which rule dropped them.

Step 5. Get Work Emails for the Shortlist Only

Enrich last, since most people in early batches get dropped. Crustdata returns business emails for the people you keep, with a verified option that returns only emails that pass a deliverability check, and thereโ€™s no charge when nothing is found. Contact data is on enterprise plans. Check each personโ€™s current job first, because people move.

For everyone scored 4, confirm their current employer and start date.
Drop anyone who started a new role in the last six months.
Get a verified business email from Crustdata for the rest.

Step 6. Export the Shortlist and Draft Outreach

Ask for a file with a reason next to each person, remove anyone already in your ATS, and have Claude draft a first message that mentions one real detail from each personโ€™s background. Claude drafts. You read and send.

Here's an export from my ATS.
Remove anyone on the shortlist who's already in it.
Give me a CSV with name, title, company, score, reason and business email.
Draft a first message to each person in my voice, under 90 words.
Don't send anything.
ellips

How The Firm Grew Sourcing Capacity 50% Without Hiring

How The Firm Grew Sourcing Capacity 50% Without Hiring

A two-partner hardware and engineering search firm. No sourcer in between.

The Firm

"In other AI recruiting tools,

out of 700 people shown, 500 aren't a fit. Crustdata shows 50 to 60 people that are great fits and that's a huge jump in quality."

The Firm

"In other AI recruiting tools, out of 700 people shown, 500 aren't a fit. Crustdata shows 50 to 60 people that are great fits and that's a huge jump in quality."


Dominique Kimbrough

Co-Founder, The Firm

50%

Sourcing capacity increase, no new hire

8โ€“10%

Candidate interest rate, vs 2โ€“3% industry average

650

Irrelevant results removed in a single search

Why AI Sourcing Tools Return the Wrong People

โœ•

The AI sourcing tool

โœ•

The recruiter seats you pay for

โœ•

The engineers you hired to build your own layer

Same result every time. A specific brief comes back as 700 names where 500 don't fit, because it reads your JD as keywords and has no judgment. The more context you give it, the worse it gets.

Save the Loop as a Claude Skill

A skill is a saved set of instructions Claude follows every time you call it, so the next role starts from your rubric and feedback rules instead of a blank chat. The Firm, a two-person recruiting agency, built a custom Claude skill on Crustdata that encodes how its founders judge candidates, and had it working within a week. In Claude Code, the Crustdata plugin comes with a ready-made candidate sourcing skill.

Turn what we did in this conversation into a reusable skill called [role]-search.
Include the rubric, the 1 to 4 scale, the feedback rules and the output columns.
Show me the SKILL.md before you save it.

Market mapping

size and map the talent pool

Sourcing

find and score candidates

Screening

research a candidate in depth

Enrichment

verified email and phone

ATS push

write to Loxo or Gem and sequence

ellips

How to Set Up Claude for Recruiting

Setup takes a few minutes and works on every Claude plan, including Free, which allows one custom connector. On Team and Enterprise plans, an owner adds the connector first.

  1. In Claude, open Customize, then Connectors, then Add custom connector.

  2. Name it Crustdata, paste https://install.crustdata.com/mcp and click Add.

  3. Open the new connector, click Connect and sign in with your Crustdata account.

Connectors are account-wide, so Crustdata then works on claude.ai, Claude Desktop and mobile. To check it worked, ask Claude โ€œHow many credits do I have?โ€

In Claude Code, run this, then type /mcp and pick crustdata to sign in:

claude mcp add --transport http crustdata https://install.crustdata.com/mcp

Or install the Crustdata plugin, which adds the same connector plus ready-made skills, and sign in with /crustdata:login.

Claude for Recruiters: Frequently Asked Questions

Recruiters use Claude to turn a job description into a scoring rubric, search for candidates, score them with a reason for each and draft outreach. The search needs a people database. With Crustdata's MCP server connected, Claude searches 1B+ people and returns scored batches you refine with feedback.
Most use it for job descriptions, outreach and interview questions. Sourcing teams connect it to live data and run the whole search in one conversation. The Firm, a two-person recruiting agency, built a Claude skill on Crustdata and grew its sourcing capacity 50% without hiring.
Both write job descriptions and outreach well, and both can connect to outside data. For sourcing, the data you connect matters more than the assistant, because either one can only rank the people its data source returns. Crustdata works in both, as a connector in Claude and as a plugin in ChatGPT.
It depends on the task. For writing, any of the main assistants works. For sourcing, the best setup is a strong assistant connected to a current people database with deep filters. Crustdata gives Claude 1B+ people, 60M+ companies and 350+ filters, retrieved live when you ask, with the database refreshed every 1 to 30 days.
Yes, if your ATS has a Claude connector or an API. Claude can then check the shortlist against people you already have and add new candidates. For other systems, export a CSV and ask Claude to remove the duplicates.
You need a Claude plan and a Crustdata account. Crustdata charges in credits. A search costs from about 0.03 credits per result, and no credits are used when a request returns nothing. Credits cost ten cents or less on enterprise plans and about thirty cents self serve.

Build your AI recruiting twin

Brief it, calibrate it, and encode your judgment into an AI clone that recruits just like you.

ellips

Crustdata is now live in Claude!

Find the best leads, candidates and companies with only a prompt.

Sign In