The Cognitive MCP Server: Drive Recruiting from Claude

What is MCP (the Model Context Protocol)?

The Model Context Protocol (MCP) is an open standard that lets AI assistants like Claude securely connect to external tools and data sources, so the assistant can take real actions - not just answer questions. An MCP server exposes a set of named tools; the assistant calls them on your instruction and reports back in plain language.

For recruiting, that means the chat window becomes a hiring console: instead of clicking through a dashboard, you tell your assistant what you want ("open a senior backend engineer role and draft the JD") and it drives The Cognitive for you.

What can you do with The Cognitive MCP server?

You can run the entire recruiting workflow from your AI assistant. A typical flow, in plain English:

• "Find payroll implementation consultants in Dallas who have moved clients off ADP" - search_external_candidates scores people against the brief, reopen_search goes deeper on the same search, and shortlist_candidate plus reveal_candidate_contact build the list with verified emails.
• "Start a sequence for the shortlist" - create_campaign, add_campaign_candidates and start_campaign_sequence run email outreach that stops when someone replies, on plans that include outreach.
• "Open a role for a Senior Data Engineer" - the assistant calls create_role, then generate_job_description drafts the JD and generate_interview_questions plus generate_evaluation_criteria build the interview and scoring rubric.
• "Invite these 40 applicants" - bulk_invite_candidates (or invite_candidate for one) sends AI interview invitations; candidates self-schedule from your slot window.
• "Score the new resumes against the must-haves" - score_candidate_resume grades each resume with evidence before anyone is interviewed.
• "Who did well this week?" - list_interviews and get_interview_feedback pull each candidate's scores and written feedback; you read the highlights in chat.
• "Advance the top three, reject the rest with feedback" - advance_candidate_stage and reject_candidate execute your decision; the AI never decides for you.
• "How is the pipeline looking?" - get_pipeline_summary returns the funnel in one answer, and get_credits tells you what's left on your plan.

Every tool the MCP server exposes

The full tool surface: 97 tools, grouped by area. These are the real tool names your assistant sees, nothing more and nothing less. Outreach, phone reveals and ATS tools follow your plan, the same as in the web app.

• Sourcing (19): search_external_candidates, get_search_run, reopen_search, shortlist_candidate, pass_candidate, remove_decision, list_tags, create_tag, set_candidate_tags, scout_role_candidates, list_role_candidates, list_sourcing_decisions, list_saved_searches, list_revealed_candidates, reveal_candidate_contact, check_reveal_status, get_sourcing_usage, get_sourcing_activity, import_candidates_to_role
• Outreach (11): get_role_outreach, list_campaigns, get_campaign, send_outreach, run_outreach_followups, pause_campaign, resume_campaign, create_campaign, add_campaign_candidates, start_campaign_sequence, sync_campaign
• Roles (12): setup_role_pipeline, list_roles, get_role, create_role, update_role, delete_role, generate_job_description, generate_interview_questions, list_evaluation_templates, create_evaluation_template, generate_evaluation_criteria, delete_evaluation_template
• Candidates (7): list_candidates, get_candidate, create_candidate, update_candidate, delete_candidate, score_candidate_resume, score_missing_resumes
• AI interviews (6): list_interviews, get_interview, invite_candidate, bulk_invite_candidates, cancel_interview, get_credits
• Feedback and stages (6): get_interview_feedback, list_feedback, get_candidate_stage_history, advance_candidate_stage, reject_candidate, send_feedback_for_review
• ATS (14): get_ats_connection, list_ats_jobs, list_ats_stages, set_ats_source_stage, set_ats_writeback, sync_from_ats, push_to_ats, get_ats_pipeline, link_ats_job, unlink_ats_job, set_ats_automation, run_ats_automation, get_ats_activity, send_to_ats
• Sheets and pipeline (4): list_sheet_connections, set_sheet_auto_sync, remove_sheet_connection, get_pipeline_summary
• Plan and billing (10): get_plan, preview_plan_change, change_plan, cancel_plan, resume_plan, change_seats, buy_credits, buy_interview_credits, list_invoices, get_team_usage
• Team and data (8): list_members, list_member_roles, add_member, change_member_role, remove_member, find_person_data, export_person_data, delete_person_data

How do you set up The Cognitive MCP server?

Setup takes about 2 minutes and is self-serve: you need a The Cognitive account (the free trial works) and nothing else.

• 1. In Claude, open Settings, then Connectors, then Add custom connector.
• 2. Paste https://app.thecognitive.io/mcp and sign in with your The Cognitive account. Claude gets access scoped to you: it sees what you see in the app, nothing more. The server also works with any MCP-compatible client.
• 3. Ask your assistant to get started - for example: "Create a role for a Senior Backend Engineer, generate the job description and interview questions, and show me the evaluation criteria."
• 4. Actions run against your real workspace using your plan's interviews and credits - the same account the dashboard uses, so everything stays in one pipeline.

Why drive recruiting through MCP?

• One interface: role setup, JD writing, question generation, invitations, feedback review, and pipeline reporting happen in the same conversation where you already work.
• Agentic recruiting with human control: the assistant executes the busywork (a role with AI-generated JD, questions, and criteria takes about 10 minutes to stand up), while stage moves and rejections only happen when you say so.
• Evidence stays attached: interview feedback pulled over MCP is the same feedback as the dashboard - a 1 to 5 score per criterion and overall written feedback, not vibes.
• Scriptable hiring: because tools are standard MCP, you can chain them - score resumes, invite everyone above the bar, and summarize the pipeline - in one instruction.

Frequently Asked Questions