10+ Hires a Day With Claude MCP

What MCP means for a recruiter

The guide skips the jargon. MCP, the Model Context Protocol, is a connector that lets Claude work inside other software instead of only giving advice. Without it, a recruiter exports a CSV from one system, uploads it to another, sends invites one at a time and chases reports across tabs. With it, a sentence in a chat window can pull candidates, send invites and fetch results.

The Cognitive offers an MCP server for exactly this, so Claude can work with your roles, candidates and interview reports from a conversation. The guide's author describes it as the moment the AI stops advising and starts operating the tools.

The setup, done once

Before the week began, the author had three connections in place: applicants flowing from the careers page into the ATS, the ATS connected to The Cognitive, and The Cognitive's MCP connected to Claude. The Cognitive connects to more than 60 ATS platforms, including Greenhouse, Lever, Ashby, Workable and Bullhorn, listed on the integrations page. For candidates imported from the ATS, it can write a note with the interview score, summary and report link back to the candidate record.

One rule came before any automation: every role gets a rubric before it gets a single applicant. The rubric is what the interview asks against and what the scores measure. The author's warning is blunt. Automating a role with no rubric just automates chaos.

The week, one prompt at a time

On Monday, the first prompt asked Claude for one table per role showing open seats, total applicants, new applicants since last week and anyone stuck at a stage for more than three days. The author says it replaced about an hour of clicking through dashboards.

Next came an audit of the early screening criteria. The prompt groups everyone screened out by reason and flags any single reason behind more than 40 percent of them. The author credits it with catching a location rule that was quietly excluding candidates willing to work remotely, just because of how they typed their city.

Then came the step readers find hardest to believe: one prompt inviting every remaining candidate across all roles to an AI interview, with Claude confirming the count before anything is sent. The author reports more than 300 invites going out this way. The Cognitive supports bulk invites, and candidates book their own slot without creating an account, so people interviewed early in the morning, late at night or at the weekend.

For the next two to three days the live AI video interviews ran without the recruiter. At the end of day three, a report prompt asked for candidates ranked by weighted score, the strongest and weakest answer for the top ten with quotes, integrity flags with timestamps, contradictions between answers, and a suggested cut line with reasoning for each role.

The human part stays human

The author then reviewed every scorecard at about a minute per candidate. That is enough, the guide argues, because the evidence is already organised: the job is to check whether the quote supports the score and whether a flag matters for this role. Where instinct disagreed with the report, the author watched that part of the recording, which happened only a handful of times all week.

Approved rejections went out with specific feedback drawn from the scorecard, strengths first. Each role ended with two or three finalists, each of whom the author spoke to for thirty minutes before sending them to the client with the recording, transcript and scorecard. In The Cognitive, nothing is rejected automatically, and the guide is clear that humans made every hiring decision.

The bonus prompt library and the lessons

The guide includes four reusable prompts that work with any tool you can export from.

• Rubric Generator: five weighted parameters, each with what a 1, 3 and 5 answer sounds like, two questions and a follow-up that catches exaggeration.
• Monday Pipeline Health Check: pass-through rates at each stage, flagging applied to screened below 20 percent or above 80 percent.
• Stuck Candidate Rescue: everyone idle at a stage for more than five days, with the pending action and who owns it.
• Client Update Draft: a short weekly note for agency and embedded clients with finalists, highlights and timing.

What's inside the download

Beyond the workflow, the full guide explains why the author's offer rate stayed high (speed, respect for candidates' time and evidence the client could inspect), gives a per role human time ledger of about 5 to 6 hours, and lists the lessons nobody mentions: the bottleneck moves to your own review time, the rubric becomes where hires are won or lost, and high volume without feedback damages your reputation at scale. It ends with a five day plan you can run without any product.

Frequently asked questions

What is Claude MCP in recruiting?

MCP is a protocol that lets Claude connect to and operate other software. In recruiting it means you can ask Claude, in plain English, to pull candidates, send interview invites or summarise reports from a connected platform instead of clicking through dashboards.

Can Claude really invite hundreds of candidates in one prompt?

The guide's author reports sending more than 300 invites with one prompt, with Claude confirming the count first. The Cognitive supports inviting any number of selected candidates at once.

Do I lose control of hiring decisions if Claude runs the workflow?

No. In the guide, the recruiter reviews every scorecard, approves every rejection and personally interviews the finalists. The Cognitive does not reject candidates automatically.

Which ATS can this workflow connect to?

The Cognitive integrates with more than 60 ATS platforms, including Greenhouse, Lever, Ashby, Workable, Bullhorn and Workday. Teams on other systems can use CSV upload and a read only report link.

Can I use these prompts without MCP?

Yes. The rubric generator works in any chat assistant, and the pipeline and stuck candidate prompts work on an ATS export pasted into Claude. MCP removes the copying, not the method.