MCP for Recruiters: How to Source Candidates From ChatGPT or Claude
Published and last updated November 12, 2026.
Quick answer: A recruiting MCP server connects an AI assistant such as ChatGPT or Claude to recruiting tools. The assistant can search candidates, open profiles and save shortlists through the connection. Wrangle provides a hosted MCP server with OAuth sign-in. SeekOut, Greenhouse, Leonar, Workable and Pinpoint also document recruiting MCP servers, with different access to sourcing data, ATS records and CRM workflows.
For a recruiter, the practical benefit is continuity. You can describe a role, inspect candidates and build a shortlist in the same conversation instead of moving results between a chat and a spreadsheet.
This guide covers setup, everyday workflows, permissions and costs. It complements our more technical guide to building a sourcing agent with the Wrangle API and MCP, which focuses on turning past hiring decisions into a recurring feedback loop.
What is an MCP server for recruiting?
An MCP server exposes tools and data to an AI assistant using the Model Context Protocol, an open standard. The assistant sends a structured request, such as a candidate search, to the server. The server runs the supported action and returns the result for the assistant to use.
The connection supplies access to a real recruiting system. Asking an unconnected chatbot to find candidates is different: it may search the web or work from information you pasted, but it cannot use your Wrangle collections or your ATS records without an authorized connection.
Recruiting MCP servers generally cover three kinds of work:
- Sourcing: finding people for an open role, reviewing profiles and enriching contact information.
- ATS workflows: working with jobs, candidates and activity already in your hiring pipeline.
- CRM workflows: managing candidate relationships, client records, projects and outreach.
These categories overlap. A sourcing server can expose saved candidate records, and an ATS server can support actions as well as reporting. Choose by the work you need the assistant to do, not just by whether a vendor lists “MCP” as a feature.
Which recruiting tools have an MCP server?
The vendors below document MCP support as of November 12, 2026. This is a comparison of published capabilities, not a hands-on test of every connection. Wrangle publishes this guide.
An ATS connection and a sourcing connection can be useful together. One can help you discover candidates; the other can help you work with records already in process. You do not need several servers just to complete a first search.
Sources: each vendor's MCP product or help documentation, checked November 12, 2026. Tool availability, plans and client support can change.
What should you check before connecting a recruiting MCP server?
Before adding a connection, answer six practical questions:
- What data can it reach? External profiles, saved candidates, ATS records and private notes are different sources.
- Which account is connected? Confirm the recruiter, organization or personal account and the credit balance being used.
- Which actions change something? Saving candidates, updating stages and launching outreach need more attention than opening a profile.
- Where does approval happen? Check both the assistant's tool permissions and the recruiting platform's confirmation steps.
- What costs credits? A tool can be read-only and still incur a usage charge. Check searches, enrichment, collection access and saves.
- How do you disconnect it? Make sure you can find the connected-client controls before you start the pilot.
Use your team's approved AI account for candidate data. Review the access requested during sign-in and keep the first workflow limited to the role and records you intend to work on.
What can a recruiter do with Wrangle's MCP server?
Wrangle exposes tools for candidate search, full profiles, contact enrichment, collections and connected sources. It also supports actions around projects, ATS export, agents and email outreach. What the assistant can do depends on the access you grant and the permissions and features available to your account.
The five examples below use one search and shortlist. They are sample instructions for an assistant, not promises about how many matching candidates or email addresses it will find.
1. Find candidates from a hiring brief
Find 25 senior backend engineers who have scaled a payments system at a Series B or C company. Use Deep mode and show the evidence for each requirement.
The assistant calls Wrangle's search tool with your criteria. Search returns ranked profiles with match scores. Deep mode uses individual candidate evaluations to determine the ranking, but the MCP response contains compact profiles and scores rather than the full evaluation notes. Ask the assistant to support its summary with profile evidence and open full profiles where needed.
Specify the role, location and experience that matter. Review the evidence before treating a match as qualified. Missing profile information should stay unknown rather than becoming an assumption in the assistant's summary.
2. Open a candidate's full profile
Open the third candidate's full profile. Summarize their last two roles and separate documented work from anything we still need to verify.
Search results contain a compact profile. The full-profile tool can retrieve more experience, education, skills and company context when available. That lets you investigate a promising candidate without treating the first search row as the complete record.
Full-profile enrichment has its own credit charge. Open the candidates you want to examine rather than automatically enriching the entire result set.
3. Save a shortlist to a collection
Save candidates 2, 5, 8 and 12 to a new collection called Backend Engineer — Client A. Use the candidates from this search.
Collections are real Wrangle records, so you can return to the shortlist in the app. Saving candidates does not automatically share them with every teammate: access depends on collection and project permissions.
Use a distinct collection for each role or client. If several collections have similar names, ask the assistant to confirm the destination before saving. Passing known candidate IDs from the search also avoids resolving the same people from profile URLs again.
4. Enrich contact details for selected candidates
Find work emails for those four shortlisted candidates. Show which addresses were found and the credits used.
Wrangle's email-enrichment tool can work with individual candidates, a batch or a collection. In collection mode, it skips stored emails and saves newly found addresses.
Enrichment is metered and an address is not guaranteed. A returned address is contact information, not evidence that the candidate is looking for a role.
5. Refine after hiring-manager feedback
The hiring manager wants production payments experience, not just payments clients. Make that explicit, search for 25 more candidates, and show the best 10 for review.
The assistant can carry the conversation's feedback into a new search. Keep the updated criteria visible so you can check what changed. A second search is another metered request, even when you only ask to see ten of its results in the chat.
By default, Wrangle's MCP search excludes profiles that the connected account previously saved or hid. If you want to reconsider those people, ask the assistant to include them. Otherwise, the same prompt may return a different set after you save a shortlist.
Can Wrangle's MCP server send outreach?
Yes. The current server can create email sequence drafts and launch them through a preview-and-confirm flow. It is not limited to drafting text for you to copy into the app.
Creating a draft is separate from enabling the sequence. The launch workflow previews the audience and settings, then requires confirmation before sending starts. Review the recipients, connected mailbox, message copy and schedule before approving. Once enabled, the sequence follows its sending settings; it does not ask again before every scheduled email.
For a first pilot, ask the assistant to stop at the draft. You can add sending later after reviewing the workflow. Client approval settings and Wrangle's access controls both matter; a sentence in a prompt is not a replacement for those controls.
How do you connect Wrangle to ChatGPT?
Use ChatGPT on the web with permission to add custom MCP servers. Workspace restrictions can affect whether the option is available. You also need a Wrangle account with access to the relevant developer surface and sufficient credits for the actions you run.
- In Wrangle's Developer area, open MCP and copy the server URL for the account or workspace you want to connect.
- In ChatGPT, open Plugins, select the plus button, then Add custom MCP server.
- Name it Wrangle and paste the URL into Server URL.
- Choose OAuth authentication. Use automatic client registration if the setup asks for a registration method; Wrangle supports dynamic registration.
- Review the connection warning, create the plugin, then install it and complete Wrangle sign-in. Check the account, billing destination and access level on Wrangle's approval screen.
- Enable the plugin in a conversation and ask it to run one search using a specific mode and result count.
The workspace server address is https://app.usewrangle.com/api/mcp/remote. Copying it from Wrangle is preferable because a personal developer connection can include an additional account-mode parameter. Preserve the full URL you copied.
These steps follow the current OpenAI custom MCP setup documentation. If your workspace hides the option, ask its administrator about custom-server access rather than switching to a personal account with company candidate data.
How do you connect Wrangle to Claude?
Claude supports remote custom connectors. Organization owners or authorized admins may need to add the connector before team members connect their own accounts.
- Copy the server URL from Wrangle's MCP tab.
- Open Claude's connector management area. Depending on the rollout, this appears under Plugins or Customize → Connectors.
- Choose Add custom connector, or Add → Custom → Web, and name it Wrangle.
- Paste the URL and use OAuth sign-in. If Claude asks how to identify its OAuth client, choose Register automatically.
- Complete Wrangle sign-in and review the account, billing destination and access requested.
- Enable the connector for the conversation and run one search.
Claude is moving connector management into its Plugins interface, so the labels can differ across accounts during the transition. The Claude remote connector guide covers setup and organization requirements.
For another client, confirm that it supports remote MCP and Wrangle's OAuth flow. Supporting MCP in general does not establish compatibility with every server or authentication method.
How do permissions, approvals and disconnection work?
OAuth and account access
OAuth lets you authorize the connection by signing in to Wrangle. You do not paste your Wrangle password into the conversation. The consent screen identifies the connector, account and billing destination, and offers Full access and Read only choices.
Choose the level that matches the workflow. Read-only access is useful for research, but it can still spend credits on supported reads and searches. Full access is needed for supported actions that change records.
Workspace and collection permissions
The assistant can use resources the connected account is permitted to access. Some list tools can show accessible projects or collections across multiple organizations, so confirm the intended project and collection instead of assuming every result belongs to the active client engagement.
Collections, ATS sources and outreach sequences can have different access rules. Seeing a record in a list does not establish permission to edit it or launch its sequence.
Disconnecting an assistant
In Wrangle's Developer area, open MCP → Connected clients and choose Disconnect for the relevant client. Also remove or disconnect the corresponding plugin or connector in the assistant.
Disconnecting access is not the same as undoing completed work. Saved collections remain saved, and a sequence or scheduled agent you already enabled needs to be paused separately if you want that work to stop.
What does Wrangle's MCP server cost?
API and MCP actions use Wrangle credits. The AI assistant's own subscription or model charges are separate. These rates describe API and MCP search, not the search allowance in the Wrangle app.
Each search requests 25–200 profiles, with 50 as the default. The requested count is a ceiling, not a guarantee. Billing uses the profiles actually returned, and the search total is rounded down to a whole credit. Asking for fewer than 25 does not create a smaller request: the minimum still applies.
Search is only part of a workflow's cost:
The full tool catalog includes additional charges for actions such as ATS export, live sequence details and AI personalization. Check the tool's cost disclosure before expanding a workflow. A read-only operation is not necessarily free, and a free tool can trigger other billable work.
For a simple example, a Fast search that returns 50 people costs 50 credits. Saving ten known candidate IDs in one call adds 1 credit. Four successful email lookups add 20 credits, for 71 credits total, before any full-profile enrichment, collection polling or other actions.
Confirm the selected billing balance in Wrangle before a pilot. Personal developer credits and workspace credits are different billing contexts.
Should you use MCP or the REST API?
MCP exposes tools to an assistant and can authenticate through OAuth. The REST API accepts structured requests authenticated with an API key. Both can operate on Wrangle data, but the available actions and account context need to match your intended workflow.
You do not need to build or host a server to use Wrangle's hosted MCP connection. Custom application logic, scheduling and reporting may still require engineering work.
Does Wrangle's MCP work with my ATS?
Wrangle integrates with many major ATS platforms. The MCP connection and the ATS connection are separate: one lets an assistant call Wrangle tools, while the other makes supported ATS data and actions available to Wrangle.
Current MCP tools can discover available connected sources and support candidate pushes to a connected ATS through a preview-and-confirm flow. Provider capabilities, permissions and the selected candidates determine which actions are available. Confirm the destination account and job before approving an export.
If you need broader ATS administration or pipeline reporting, your ATS vendor's own MCP server may be useful alongside Wrangle. Connecting Wrangle does not automatically grant every action available in another system.
How do you pilot a recruiting MCP server in one week?
- Choose one role. Write the brief as you would explain it to a sourcer, including the requirements that are hard to capture in a title.
- Connect one assistant. Confirm the Wrangle account, access level and credit balance before searching.
- Run one bounded search. Use Fast mode with 50 requested profiles, or Deep for additional evaluation before ranking. Note the mode and charge.
- Review and save a shortlist. Choose the candidates worth discussing and save them to one clearly named collection.
- Apply specific feedback. Revise one or two criteria and compare the next search with the first.
- Enrich selectively. Look up contact details for candidates you intend to contact, and review any outreach draft before enabling it.
- Compare the work. Record qualified candidates, time spent checking results and total credits used. Decide whether the connection improves your normal process.
Keep a short criteria document outside the conversation if other recruiters will repeat the workflow. It gives the next person a clear starting brief and makes changes easier to review.
Frequently asked questions
What is the best MCP server for candidate sourcing?
Choose one that exposes candidate search, not just records already in an ATS. Wrangle and SeekOut document sourcing capabilities. Greenhouse, Workable and Pinpoint focus on work inside their hiring systems. Test the sources and actions you need rather than choosing by tool count.
Does Wrangle have an MCP server for ChatGPT and Claude?
Yes. Wrangle hosts a remote MCP server with OAuth authentication. Use a client that supports its connection requirements, and confirm that your account or workspace permits custom connectors.
Do I need an engineer to set it up?
Not for the hosted connection. Setup involves adding the server URL and signing in to Wrangle. An administrator may need to approve access in your AI workspace. Custom integrations built with the REST API are a separate project.
Does MCP search return every matching candidate?
No. Wrangle returns a ranked, bounded set of results. Requests allow 25–200 profiles, with 50 by default, and fewer may be returned when there are not enough matches.
How much does a search cost?
A search returning 50 profiles costs 25 credits in Turbo, 50 in Fast, 100 in Normal or 150 in Deep. Enrichment, saves and some reads have separate charges. The assistant's own subscription is billed separately.
Can the assistant send emails through Wrangle?
Yes, with the required access and setup. MCP can create sequence drafts and enable them after a launch preview and confirmation. Once enabled, sending follows the sequence's settings. Keep the workflow at draft stage if you only want copy for review.
Can I stop an assistant from accessing Wrangle?
Yes. Use Disconnect under MCP → Connected clients in Wrangle's Developer area, then disconnect the corresponding client-side connection. Pause any enabled sequences or scheduled work separately; removing access does not reverse prior actions.
Is MCP the same as an ATS integration?
No. MCP connects an assistant to Wrangle tools. An ATS integration connects Wrangle with a supported hiring system. A connected ATS may supply data or actions to the MCP tools, but permissions and provider support still apply.
Connect Wrangle to your assistant
Use the Wrangle developer guide to review setup and tool costs, then try one real brief. Search, inspect a few profiles and save a shortlist before adding more actions.
