AI Recruiting Agent: Find and Reach Candidates with MCP

A recruiting sourcing pipeline can span LinkedIn Recruiter, a CSV, an email finder, and an outreach platform. An AI recruiting agent connected to B2B data via MCP can coordinate search, narrowing, contact reveal attempts, and outreach drafting in one conversation. Describe who you need, review matches from 700M+ professional profiles, and request available contact fields only for the people you choose.

The sourcing pipeline nobody loves

A manual version of the sourcing workflow can look like this:

1. Search LinkedIn Recruiter. Set your filters — title, location, years of experience, industry. Scroll through results. Open profiles that look promising. Maybe save them to a project.
2. Move the shortlist into another sourcing tool. Depending on your stack, that may mean a supported CRM sync or a CSV export from a separate sourcing product. Some records may be incomplete or out of date.
3. Run emails through a finder tool. Upload your CSV to an enrichment provider and wait for the results. Some records may return contact fields, while others may return no available data. Pricing and field entitlements vary by provider.
4. Clean the data. Remove duplicates. Review candidates who may have changed jobs since the source record was captured. Fix formatting issues from the export.
5. Import into your ATS or outreach tool. Map the CSV columns to your system’s fields and resolve any import errors.
6. Research candidates for personalization. Open each person’s LinkedIn profile again (you already looked at it in step 1, but the data didn’t carry over). Note their recent projects, mutual connections, career trajectory. This is where the good outreach comes from, and it’s entirely manual.
7. Write outreach messages. Draft and review personalized emails or InMails based on your research.

Each step is a context switch. Each transition introduces delay and data loss. The candidate’s information exists in your LinkedIn tab, your CSV, your email finder dashboard, and your ATS — four versions of the truth, none of them perfectly in sync.

Roles and contact details change over time, so any result should be treated as a point-in-time record and reviewed before it enters a recruiting workflow.

What an AI recruiting agent looks like in practice

An AI recruiting agent can coordinate several sourcing stages in one conversation. Connected to a B2B data provider through MCP (Model Context Protocol), it can search professional profiles, make reveal attempts, and use returned fields in the same session.

Here’s what the workflow actually looks like. You tell your Claude Code agent:

The agent uses call_operation with operationId: peopleSearch and supported filters:

{
  "tool": "call_operation",
  "arguments": {
    "operationId": "peopleSearch",
    "params": {
      "body": {
        "searchParams": {
          "jobTitleV3": {
            "include": [{
              "type": "functional",
              "seniority": ["senior"],
              "keywords": ["backend", "software engineering"]
            }]
          },
          "industry": {
            "anyOf": ["Financial Services"]
          },
          "location": { "include": ["Berlin, Germany"] },
          "jobStatus": { "status": "currently-employed" }
        },
        "pageSize": 25
      }
    }
  }
}

The agent returns matching profiles with the fields available in the search response, such as name, title, company, and location. Search costs 0 credits. You can review those fields, refine supported filters, and select candidates for reveal attempts.

Then you say:

The agent can use the standard reveal operation for each selected candidate:

{
  "tool": "call_operation",
  "arguments": {
    "operationId": "syncQuickContactReveal",
    "params": {
      "body": {
        "linkedinUrl": "https://www.linkedin.com/in/example-profile",
        "enrichmentType": {
          "getWorkEmails": true,
          "getPersonalEmails": false,
          "getPhoneNumbers": true
        },
        "validateEmails": true
      }
    }
  }
}

Each reveal attempt costs 1 credit, including not-found results. When available, work email and direct phone are returned together under that credit. Five selected candidates can therefore mean five reveal attempts and five credits, whether each attempt returns contact fields or a not-found result.

Lite uses a lower-yield lookup. Standard balances speed and coverage. Turbo prioritizes speed, while Exhaustive trades latency for the broadest available provider coverage.

The agent can combine the profile context and any contact fields that were returned, then help draft personalized outreach:

The agent can draft a message for each selected candidate using the profile details available in the session. A recruiter should review the matches, contact fields, and final messages before sending. The benefit is a connected workflow without a CSV handoff; completion time depends on the client, reveal mode, provider responses, and the recruiter’s review.

Before and after: the workflow shape

The useful comparison is the number of handoffs, not a promised completion time. Actual duration depends on the search, reveal mode, provider responses, and human review.

The agent workflow reduces tool handoffs by coordinating search, narrowing, reveal attempts, and drafting in one place. It does not eliminate the recruiter’s responsibility to assess fit, confirm returned fields, and approve every message.

Real search queries for common recruiting scenarios

Here are practical requests an agent can send through call_operation with the supported peopleSearch and paginatedCombinedSearch operations. You describe what you need in plain English and the agent constructs the matching operation call.

Engineering leadership at mid-sized software companies

{
  "tool": "call_operation",
  "arguments": {
    "operationId": "paginatedCombinedSearch",
    "params": {
      "body": {
        "companyConfig": {
          "searchParams": {
            "industriesV2": { "anyOf": ["Software"] },
            "employeeCountV2": {
              "lowerBoundExclusive": 200,
              "upperBoundInclusive": 500
            }
          },
          "pageSize": 25
        },
        "profileConfig": {
          "searchParams": {
            "jobTitleV3": {
              "anyOf": [{
                "type": "functional",
                "seniority": ["vp", "director"],
                "keywords": ["engineering"]
              }]
            },
            "jobStatus": { "status": "currently-employed" }
          },
          "pageSize": 25
        }
      }
    }
  }
}

This searches for engineering leaders at software companies in the 201–500 employee range. Review the returned company and role details before inferring hiring stage or open headcount.

Enterprise sales reps (for recruiting, not selling to)

{
  "tool": "call_operation",
  "arguments": {
    "operationId": "peopleSearch",
    "params": {
      "body": {
        "searchParams": {
          "jobTitleV3": {
            "include": [{
              "type": "functional",
              "seniority": ["senior"],
              "keywords": ["enterprise", "account executive"]
            }]
          },
          "industry": { "anyOf": ["Software Development"] },
          "country3LetterCode": { "anyOf": ["USA"] },
          "jobStatus": { "status": "currently-employed" }
        },
        "pageSize": 25
      }
    }
  }
}

Useful when you’re hiring enterprise AEs and want to find people who are already doing the job at software companies. Filter by seniority to skip SDRs and junior reps.

Product management leadership

{
  "tool": "call_operation",
  "arguments": {
    "operationId": "peopleSearch",
    "params": {
      "body": {
        "searchParams": {
          "jobTitleV3": {
            "include": [{
              "type": "functional",
              "seniority": ["senior", "manager", "director"],
              "keywords": ["product"]
            }]
          },
          "jobStatus": { "status": "currently-employed" }
        },
        "pageSize": 25
      }
    }
  }
}

This example omits location and industry filters to cast a wider net. You can narrow the next search after reviewing the initial results.

Data engineers at healthcare companies

{
  "tool": "call_operation",
  "arguments": {
    "operationId": "peopleSearch",
    "params": {
      "body": {
        "searchParams": {
          "jobTitleV3": {
            "include": [{
              "type": "plain",
              "term": "data engineer"
            }]
          },
          "industry": {
            "anyOf": ["Hospitals and Health Care"]
          },
          "country3LetterCode": { "anyOf": ["USA"] },
          "jobStatus": { "status": "currently-employed" }
        },
        "pageSize": 25
      }
    }
  }
}

Domain expertise matters in healthcare data. This query narrows the search by role and company industry; it does not prove a person’s familiarity with HIPAA, HL7, or any other domain requirement. Check each profile and confirm that experience during recruiting.

Responsible recruiting still matters

Any conversation about AI-powered recruiting needs to address data use directly. Descovo can return available work email and direct phone fields through a reveal attempt; availability differs by person and attempt.

Your organization remains responsible for deciding whether and how to use a returned field, following applicable law, honoring opt-out requests, and applying its own recruiting and data-retention policies. A contact result is not consent to send irrelevant or indiscriminate outreach.

In practice:

• Reveal selectively. Search and narrowing cost 0 credits, so review profile fit before spending a credit on a reveal attempt.
• Expect partial results. An attempt may return a work email, a direct phone, both fields, or a not-found result.
• Review before use. Confirm that the profile, returned fields, and message are appropriate for the role and your organization’s policies.
• Keep a human accountable. The agent can coordinate data work, but a recruiter should own candidate selection and communication.

The point of an AI recruiting agent is to reduce manual handoffs, not to remove judgment. Make reveal attempts only for people you would genuinely want to approach about a specific role, and keep the outreach relevant and reviewable.

What Descovo does not replace in your recruiting stack

Honesty about scope: Descovo is a data layer for searching people and attempting to reveal available contact fields. It is not the rest of your recruiting stack.

Things Descovo does not do:

• It is not an ATS. It doesn’t track candidates through your hiring pipeline. Keep using Greenhouse, Lever, Ashby, or whatever your team runs on.
• It does not send emails or InMails. When a work email is available, it can return that field. Sending the message remains your outreach tool’s job — or your own Gmail.
• It does not score or assess candidates. No skills assessments, no culture-fit predictions, no AI-generated rankings of who you should hire. That’s a different (and more fraught) problem.
• It does not replace LinkedIn. LinkedIn supports professional profiles and relationships. Descovo supports B2B search and contact reveal attempts. They can play different roles in the same recruiting workflow.

The right mental model: Descovo is a data engine that can feed your existing workflow. Your agent uses it to search and request contact fields. Your ATS tracks the pipeline. Your outreach tool sends the messages.

The recruiting function is splitting in two

The opportunity is not to remove the recruiter. It is to let an agent coordinate repetitive search and data steps so the recruiter can spend more attention on fit and communication.

Recruiting is becoming two distinct skills:

1. Defining what “great” looks like for a role. This is judgment work. It requires understanding the team, the culture, the technical requirements, the growth trajectory of the company, and the market for the talent you’re hiring. Search filters can support that judgment, but they do not define it.
2. Building relationships with those people. Once you know who you want, recruiting becomes relationship work. Understanding their motivations. Selling the opportunity in terms that resonate with their specific career goals. Navigating competing offers. Closing. This is deeply human work that gets better with experience and empathy.

Everything in between — the searching, the data gathering, the list building, the CSV wrangling, the email finding, the basic research — is agent work. It’s the kind of repetitive, multi-tool, data-fetching work that AI agents are specifically good at.

Recruiters can use this split to spend more attention on the parts of recruiting that require human judgment — evaluating fit and building rapport — and less on moving data between tools.

An agent is most useful when it reduces a repeatable handoff while keeping selection, review, and outreach decisions with a person.

Key takeaways

• A manual recruiting pipeline can span search, CSV export, enrichment, cleaning, import, research, and outreach. An MCP agent can coordinate search, narrowing, reveal attempts, and drafting in one session.
• Within a compatible agent, the Descovo search-to-reveal portion can run in one conversation. Moving reviewed results into an ATS or outreach tool remains a separate downstream step.
• Search and narrowing cost 0 credits. Each reveal attempt costs 1 credit, including not-found results. When available, work email and direct phone are returned together under that credit.
• Contact field availability varies. Review each result and follow your organization’s legal, privacy, recruiting, and retention policies before using it.
• Descovo is the data layer, not the whole stack. Pair it with your existing ATS, outreach tools, and LinkedIn for a complete workflow.
• Agents can handle repeatable data work while recruiters retain responsibility for judgment calls and relationship building.