AINEWS
CrustRecruiter puts Claude in the recruiter's chair
Crustdata's new MCP skills let a model run a recruiting workflow against the company's people dataset. The open question is what 'judgment' means when the same shortlist reaches every user.
Crustdata's new MCP skills let a model run a recruiting workflow against the company's people dataset. The open question is what 'judgment' means when the same shortlist reaches every user.
The line that cracks CrustRecruiter's pitch open is on the launch page itself. "Claude handles reasoning and memory; Crustdata handles the manual sourcing work," the Product Hunt listing reads. It is a clean division of labor, the kind of partnership framing that makes AI-agent products feel additive rather than displacing. The follow-on is the sentence that should give a hiring manager pause: when the same five skills run against the same 800M-profile pool for every recruiter who installs them, what exactly is the recruiter's judgment applied to?
CrustRecruiter launched on Product Hunt on June 11, 2026, billed as a way to "turn Claude into a recruiter." Mechanically, it is a set of skills installed into Anthropic's Claude through the Model Context Protocol (MCP) that give the model live access to Crustdata's people and company data, plus five recruiting-oriented skills per the launch page. Crustdata, a Y Combinator-backed B2B data provider, describes itself on its corporate homepage as offering real-time people and company data through API, bulk CSV feeds, and MCP for AI agents. The product is the recruiting-shaped instance of a broader mechanism the company has been building: B2B people and company data flowing to AI agents via MCP rather than through SaaS dashboards or REST endpoints.
The "judgment split" is the maker's thesis, and it is worth taking seriously precisely because it is the claim that does the most work in the pitch. The argument is that human recruiters were never doing the work that MCP automation now handles: pulling titles, employer histories, contact details, and tenure out of profiles and into shortlists. That grinding moves into the model. The reasoning, the memory, the editorial calls about which candidates to advance, that is supposed to stay with the human. It is a respectable claim about where value lives in a hiring workflow. It is also a claim the same launch page does not actually demonstrate. The same five skills, run against the same pool, will tend to surface the same candidates for the same job description, and the same shortlist going to two different recruiters with two different tastes is no longer a feature of the tool, it is a question for the buyer.
That is the question worth holding open, and the launch page does not close it. CrustRecruiter is the recruiting-specific instance of a data layer that, according to Crustdata's own site, already powers six named verticals: AI SDRs, sales automation, investing platforms, internal sales, PE/M&A, and recruiting. The pattern is consistent across them. The model gets a live data pipe; the model is told which skills it can call; the human is positioned as the editor at the top of the loop. It is a defensible architecture, and one an outside observer can describe without buying or rejecting the marketing.
What the launch does not have, yet, is much independent evidence. The Product Hunt page lists 2,300 followers, a 5.0 rating, and a single review. A 5.0 rating on one review is a data point, not a market signal, and the Product Hunt page is also where the only substantive critical signal on the record sits: that lone review flags pricing as a problem for startups and small businesses. The customer references on crustdata.com, the CTO at Perplexity, the head of marketing at Anthropic, the lead project manager at GitHub, are marketing-grade, not audited adoption. The "800M+ candidate profiles" figure on the launch excerpt does not match the "250M+ people profiles" figure on Crustdata's site; both are company-controlled, and the gap is the kind of thing a buyer should ask about before signing.
What to watch. The first independent recruiter to publish a real workflow with CrustRecruiter, and the first sourcing competitor to publish a side-by-side, will move this story from launch into evidence. Until then, the most useful framing is not "is this the future of recruiting" but a narrower, sourceable one: the recruiter is being repositioned as the editor of an AI shortlist, and the open question is what that editor's judgment is actually applied to, when the shortlist is being produced by a model the editor did not configure, against a pool the editor did not curate, using skills the editor did not write.