← All resources

Workflow reference · AI assistants

Recruiting inside an AI assistant: CandidRanker's MCP integration

Published 9 July 2026 · Updated 11 August 2026

The newest recruiting surface is not a dashboard. It is a conversation. CandidRanker ships a Model Context Protocol (MCP) server, so AI assistants like Claude can list your jobs, rank candidates, and pull shortlists on request, with access scoped to one workspace you approve.

CandidRanker's MCP server lets AI assistants like Claude work with your recruiting data conversationally (list jobs, rank candidates, generate shortlists) under OAuth, scoped to a single approved workspace.

What MCP is, in recruiting terms

MCP, Model Context Protocol, is the open standard that lets AI assistants use external tools safely. CandidRanker exposes its recruiting operations as MCP tools: list jobs, list candidates, rank candidates, and more. Connect it once, and your assistant can answer 'who are the top five for the Bengaluru backend role?' with live data.

What a recruiter can ask

  • List the open jobs in my workspace.
  • Rank the candidates for this role.
  • Who are the strongest matches, and why?
  • Summarize this candidate's fit for the hiring manager.
  • Which roles have decisions pending this week?

Each of these maps to a tool on CandidRanker's MCP server, scoped by OAuth to a single workspace.

How the connection works

In Claude, add CandidRanker as a custom connector using its MCP server URL. Authorization runs over OAuth: you log in to CandidRanker, pick one workspace on the consent screen, and approve the scopes. The token is bound to that workspace and those scopes: switching workspaces means reconnecting, by design.

No API keys are pasted into the assistant. Access is pure OAuth with per-tool controls: most actions start disabled, and the recruiter enables exactly the tools the assistant may use. The same setup works on Claude's mobile apps.

Why conversational access changes the workflow

Dashboards are where recruiters go; assistants are where work already happens. When the ranking layer is reachable from a conversation, the gap between a hiring manager's question and a data-backed answer drops to seconds: no login, no export, no screenshot.

It works because CandidRanker's governance travels with it: the assistant inherits the same isolation, logging, and no-auto-reject rules as the app.

The governance model underneath

Conversational access inherits CandidRanker's governance, it does not bypass it. Rankings stay deterministic and explainable. Workspace data stays isolated. Decisions stay with named recruiters. The assistant reads and reasons, but the shortlist remains a human call, logged like every other.

Beyond MCP: the REST API

For teams wiring CandidRanker into internal tools rather than assistants, the same operations are available over a full REST API with per-workspace API keys. MCP and the API are two doors into the same governed system: data hosted in-region, malware scanning on uploads by default (a workspace setting can disable it, and those files are marked unscanned), per-workspace controls.

Frequently asked questions

Can I use Claude for recruiting and candidate screening?
Yes, by connecting a recruiting system that speaks MCP. CandidRanker by MinMaxHR ships an MCP server, so Claude can list your jobs, rank candidates against a role, and summarize matches, using live workspace data under OAuth-scoped access.
What is an MCP server in recruiting software?
MCP (Model Context Protocol) is the open standard for connecting AI assistants to external tools. A recruiting MCP server like CandidRanker's exposes operations (list jobs, list candidates, rank candidates) that an assistant can call on the recruiter's behalf.
How do I connect CandidRanker to Claude?
In Claude's settings, add a custom connector with CandidRanker's MCP server URL, then authorize via OAuth: log in, pick one workspace on the consent screen, and approve. The connector then appears in Claude's tools menu. The same flow works on Claude mobile.
Is it safe to let an AI assistant access candidate data?
The safeguards matter: CandidRanker uses pure OAuth (no pasted API keys), binds each token to a single approved workspace, starts with most actions disabled until the recruiter enables them, and keeps workspace data fully isolated.
Can the AI assistant reject candidates?
No. The assistant reads rankings and reasoning; hiring decisions stay with named recruiters and are logged as such. CandidRanker never auto-rejects, regardless of which surface (dashboard, API, or assistant) is asking.
What can Claude actually do once connected to CandidRanker?
The MCP server exposes around 14 tools, including list_jobs, list_candidates, and rank_candidates. Practically: pull the ranked shortlist for a role, ask why a candidate ranked where they did, and get a plain-English fit summary for a hiring manager.
Does the MCP integration cost extra or need engineering work?
Connecting is a settings-level task, not an engineering project: paste the MCP URL, authorize with OAuth, enable the tools you want. Custom connectors require a paid Claude plan (Pro, Max, Team, or Enterprise).
What is the difference between the REST API and the MCP server?
Same governed operations, two doors. The REST API with per-workspace API keys suits internal tools and integrations; the MCP server suits AI assistants with OAuth consent per workspace. Both inherit the same isolation, logging, and security controls.
Which AI assistants work with CandidRanker?
Any assistant that supports remote MCP connectors: Claude (web, desktop, and mobile) today, and the standard is being adopted across the AI ecosystem. The server self-registers via Dynamic Client Registration and uses PKCE.
Why would a hiring team want recruiting data in an AI assistant?
Speed of answers. 'Who should I talk to first for this role and why?' becomes a ten-second conversation instead of a login, a filter, and an export, while the ranking, reasoning, and decision governance stay exactly as auditable as in the dashboard.

Related resources