Connect candidates with jobs and employers with talent

Routes agent requests to WorkorAI, a talent marketplace exposed through an MCP server (streamable HTTP at workorai.com/mcp, listed on the official MCP Registry as io.github.work0r-ai/workorai), across its dual-role tool surface - 9 candidate.* tools (job search, job detail, applications, apply, invitations, saved jobs) and a set of employer.* tools (job lifecycle, candidate discovery, invitations, applicant review).

claude mcp add --transport http workorai https://workorai.com/mcp

If the user has no API key yet, the request_access tool is called and its returned onboarding flow followed, rather than asking the user to paste credentials into chat. Requests are routed by detecting whether they're a candidate flow or an employer flow: candidate tools cover search_jobs, get_job, apply_to_job, get_applications, accept_invitation/decline_invitation, withdraw_application, and saved-job management; employer tools cover the job lifecycle (create_job -> publish_job -> close_job/archive_job), candidate discovery (search_candidates_for_job or search_candidates_by_query), and pipeline work (invite_candidate, list_applicants, get_applicant_detail, set_review_status).

Employer candidate discovery returns tiered rankings (best/good/weak) with a white-box match explanation per candidate - fit score, interview-proven skills, gaps, and a quotable rationale - instead of a black-box score. When presenting results, the tier structure and each candidate's match explanation are surfaced directly; for deeper comparison, per-candidate interview evidence is available via employer.get_candidate_evidence and employer.get_applicant_transcript.

Mutating actions (apply, invite, publish, close, delete) require explicit user confirmation before executing, since they're visible, stateful marketplace actions. Fit scores and ranks are only ever reported as returned by the tools, never fabricated, and bulk apply/invite actions are never sent without explicit approval. All operations run through the remote WorkorAI MCP server over HTTPS; the skill itself runs no shell commands, and API keys are treated as secrets stored in MCP client configuration, never in chat transcripts or committed files.

Use it when a user wants to find a job, search vacancies, apply to a position, or track applications, or when an employer wants to post/publish/update/close/archive a job, find/rank/compare/evaluate candidates, or ask why a candidate matches a role - also for setting up or troubleshooting the WorkorAI MCP connection and API key onboarding. Requires a valid WorkorAI account and API key; tools fail without one.

Input is a natural-language job-search or hiring request, routed to the matching candidate or employer tool. Output is job listings, application status, or tiered candidate rankings with white-box match explanations - each action requiring confirmation before any state-changing call.

Connects exclusively through the WorkorAI MCP server (streamable HTTP, listed on the official MCP Registry), with a full reference implementation available at work0r-ai/agent-kit (npm: @workorai/agent-kit).

Job seekers and employers using an AI agent to search/apply for jobs or find/rank/evaluate candidates on WorkorAI, who want transparent, explainable match results rather than an opaque score.