This is an engineering preview, not a marketing one. Siadhal Magos and Nolan Church sat down for a "What we're cooking" segment and showed three things in plain view: a Metaview MCP that lets you pull live ATS and interview data into Claude Code, a sourcing agent that searches your own conversations before it touches the public web, and an executive-search dashboard a non-technical recruiter built in 45 minutes of vibe-coding. None of it was theory.

The thread tying them together is the move recruiting work is making this year. The recruiter who wins is no longer the person with the best Boolean string or the cleanest spreadsheet. It is the person who can engineer the context well enough that a few agents can run real work in parallel, then judge what comes back.

That sounds abstract until you watch a sourcing agent reason out loud about which AEs hit 120 percent of quota at WeWork and Salesforce because it heard them say it on a recruiter screen six months ago. The data was always there. What changed is the protocol for getting at it and the tooling on top.

What MCP actually changes for recruiters

MCP stands for Model Context Protocol. Anthropic released it as an open standard, and the practical effect is that the agent in your chat window can pull live data from any system that publishes an MCP server, without the usual copy-paste-into-prompt workaround.

Until recently, working with Metaview data inside Claude meant exporting a transcript, pasting it in, and asking your question. Now you can ask the question directly. "Pull every senior engineer I screened in Q1 and give me the three with the strongest design-review answers" works as a single instruction. The agent fetches what it needs from Metaview and reasons on top of it inside the same session.

Siadhal was clear that this was not a roadmap moment. "It's not that deep, bro," is how he put it. One Metaview engineer shipped the MCP server because Claude Code launched, the team noticed the friction of not being able to pull Metaview data in directly, and the protocol was already there to build against. The bigger point is that the same logic applies inside any recruiting team. The valuable internal tools that used to require a sprint now live behind a prompt.

Sourcing from your own conversations first

The demo that landed hardest was the sourcing agent. The new behavior is that you can scope it to your own data before it ever touches the web. "I don't want you to look at LinkedIn. Look in my ATS, look in my Metaview instance, and find people from inside that almost like a mini database." The agent then plans, reasons, and pulls candidates back with citations to the specific conversation where it learned what it needs to know.

The example Siadhal ran live was a senior AE search: "surpassed their quota by 110 or 120 percent in the last 12 months, Bay Area, minimum 5 years experience." Half of that criteria does not appear on a resume. It does appear in the recruiter screen the AE did six months ago. The agent reads its own reasoning trail out loud, builds an Ideal Candidate Profile on the fly, decides whether to look in transcripts or scorecards or Metaview summaries, and surfaces candidates with the citation attached: "This candidate hit 120 percent of sales targets at WeWork as an inside sales rep."

The second-order effect is what Siadhal called the antidote to the AI-bias fear. Every recommendation comes with a full audit trail of why the agent decided what it decided. You can scroll back through the reasoning and challenge it. That is a meaningfully different shape than the black-box ranker most teams imagine when they hear "AI sourcing."

1

2

3

1. 1The agent writes its own plan: build an ICP, then search transcripts, scorecards, and Metaview summaries before resumes.
2. 2Reasoning trail explains why each candidate matched. "Achieved 120 percent of sales targets at WeWork" is pulled from a recruiter conversation, not a resume.
3. 3Surfaced candidates carry the citation back to the conversation that produced the signal. The audit trail is the antidote to bias fear.

Most teams default to LinkedIn the moment a new role opens. The agent reframes that habit. The richest data on whether someone is worth a conversation is the data you already paid to collect: every prior screen, every closed-won, every "this one was great but the timing was off." Reading it back at search-time is a capability that did not exist eighteen months ago.

Claude Code as the non-technical recruiter's build tool

Nolan ran the other half of the demo. He is a recruiter. He calls himself "the most non-technical person of all time." He used Claude Code, in 45 minutes, to build an executive-search dashboard with a pipeline view, a hiring-manager-call view, a weekly conversation log, and a "back burner" tab. The dashboard pulls from his ATS and the conversation log so revisionist history about which candidate said what is no longer possible.

The point is not the dashboard. The point is what 45 minutes used to buy you. Six months ago a recruiter wanting an internal tool either filed a ticket and waited, or built nothing and lived with the friction. Now the recruiter writes a description of what they want, talks Claude Code through the data sources, and watches a working prototype assemble itself. The cost of an internal recruiting tool collapsed from "engineering sprint" to "Tuesday afternoon."

This is where MCP closes the loop. The dashboard Nolan built only became powerful when it could read live data from Metaview alongside the ATS. Without the protocol, the dashboard is a static view. With it, the dashboard is the front end of the same context the agents use when they reason about candidates.

The ICP as the spine of the whole workflow

The strongest structural idea in the conversation was that the Ideal Candidate Profile, the description of what "good" actually looks like for a role, is becoming the single artifact that the rest of the recruiting workflow hangs off of.

The sourcing agent builds an ICP before it goes looking. The same ICP feeds Application Review, which runs the inbound pile against it and tells the recruiter which of the 800 applications are worth ten minutes. The same ICP can be applied to a referral pipeline. One artifact, one source of truth, every channel running against the same definition of good.

This is a real shift. Traditionally each pipeline has its own logic. Outbound recruiters look for one shape of candidate. Inbound recruiters score against a different rubric. Referrals get a thumbs-up because they came from a trusted source. A unified ICP means the same intelligence is making the same call across all three, with the same audit trail and the same compliance posture.

Why recruiter-as-context-engineer beats recruiter-as-clicker

The framing Siadhal kept coming back to is that the primary work product of a senior recruiter is now context engineering. The ICP is the artifact. The skill is the ability to describe what "good" looks like in enough detail that an agent can act on it with the same judgment a senior recruiter would.

That is a meaningfully different job description than "talented person who can run a Boolean string and a phone screen." It is closer to the skill set of a senior PM or a hiring manager who has interviewed enough people to know exactly what separates a 110-percent AE from a 90-percent one. Recruiters who can do that have just had their use multiplied. Recruiters who can't are competing with people who can spin up dashboards in an afternoon.

The same logic applies to hiring managers. The recruiter-hiring-manager partnership becomes a context-engineering partnership. The two of you sit down, you align on the ICP, the agent runs against it everywhere relevant, and the conversation in your next 1:1 is about results and edge cases, not about whether the recruiter "understands the role."

Where AI gives recruiting teams use

Sourcing

Reasons across your conversations, ATS, and Metaview instance before the public web, with a full audit trail on every recommendation.

Application Review

Runs the inbound pile against the same ICP your sourcing agent uses, so every channel feeds one consistent definition of good.

Notes

Captures every interview verbatim and structures it, so the data your sourcing agent reasons over six months later is actually there.

Reports

Surfaces hire quality, attrition, and pipeline metrics on top of the same context, so the loop closes back into the next ICP iteration.

The data backs this up. According to Metaview's 2026 AI & Hiring Alignment Report, surveying 505 recruiting leaders and hiring managers across North America and EMEA, the teams putting AI at the core of hiring are pulling away from the ones that aren't. The gap is most pronounced on the metrics that decide whether your next quarter goes well: hire quality, retention, and how often a qualified candidate slips to a competitor while you were still scheduling a debrief.

Read the full AI & Hiring Alignment Report for the rest of the breakdown by AI maturity and team type. The pattern is consistent: the teams that built shared context win on both speed and quality, and the gap is widening.

The operating shift

One: treat your interview history as a sourcing asset, not a compliance artifact. Every conversation a recruiter has is a structured signal about who can do the job. If it lives in Metaview's Notes product, the sourcing agent can read it back in six months. If it lives in a Google Doc, it is gone.

Two: build the ICP once and let it run everywhere. One description of what good looks like, used by outbound, inbound, and referral pipelines through the same agent. The skill is writing an ICP precise enough that the agent makes the same call you would. The deliverable is consistency across channels that used to operate on different rubrics.

Three: stop waiting for engineering on the small things. The dashboard, the weekly hiring-manager view, the bespoke filter against a candidate dataset are 45-minute Claude Code jobs now, not Q3 roadmap items. If a recruiter on your team is good at describing what they want, they can ship it themselves.

Four: ask whether your stack supports the protocol layer or fights it. MCP means the agent in your chat window can talk to your sourcing, your application review, your notes, and your reports as one connected system. If your point tools don't publish to the protocol, the future this episode previewed routes around them. That is the bet behind the platform approach, and the reason teams using AI broadly in hiring are 3.8x more likely to rate their working relationships as excellent.