Integration Guide

MCP Server

Run your entire hiring pipeline from the AI agent you already use. InterviewLM exposes its hiring surface as a remote Model Context Protocol server — 37 permissioned tools any MCP-capable agent can call on your behalf, in plain language.

Overview

The Model Context Protocol (MCP) is an open standard that lets AI agents securely connect to external tools and data. InterviewLM hosts a remote MCP server, so agents like Claude Code, Cursor, Codex, GitHub Copilot, Antigravity, Claude Desktop, and ChatGPT can draft pipelines, invite candidates, read evaluations, and — only if you grant it — record hiring decisions. No API keys, no new dashboard to learn.

Remote endpoint

https://interviewlm.com/api/mcp

stdio helper

npx -y @interviewlm/mcp

npm: @interviewlm/mcp — for stdio-only clients.

Connect your agent

Pick your agent and apply the one-line setup below. Native-remote clients connect to the hosted endpoint and run the OAuth 2.1 + PKCE consent in your browser; stdio-only clients use the @interviewlm/mcp helper, which handles OAuth on first run.

Claude Code

Terminal

Agent docs →

Native remote connector — Claude Code runs the OAuth consent for you.

Run in your terminal

claude mcp add --transport http interviewlm https://interviewlm.com/api/mcp

Codex

Terminal

Agent docs →

stdio via the @interviewlm/mcp helper — OAuth opens in your browser on first run.

Run in your terminal

codex mcp add interviewlm -- npx -y @interviewlm/mcp

GitHub Copilot

Terminal

Agent docs →

Native remote connector in VS Code — Copilot handles OAuth.

Run in your terminal (VS Code CLI)

code --add-mcp '{"name":"interviewlm","type":"http","url":"https://interviewlm.com/api/mcp"}'

Cursor

Config file

Agent docs →

Remote server — Cursor runs the OAuth consent in-browser.

Add to .cursor/mcp.json

{
  "mcpServers": {
    "interviewlm": {
      "url": "https://interviewlm.com/api/mcp"
    }
  }
}

Antigravity

Config file

stdio via the @interviewlm/mcp helper.

Settings → Customizations → Open MCP Config (mcp_config.json)

{
  "mcpServers": {
    "interviewlm": {
      "command": "npx",
      "args": ["-y", "@interviewlm/mcp"]
    }
  }
}

Claude Desktop

Connector

Add as a custom connector — works in Claude Desktop and claude.ai.

Settings → Connectors → Add custom connector

https://interviewlm.com/api/mcp

ChatGPT

Connector

Add as a custom connector (Plus, Pro, Business & Enterprise).

Settings → Connectors → Add custom connector (Developer mode)

https://interviewlm.com/api/mcp

How it works

Connect

No API keys or manual token setup. Add the InterviewLM connector URL (or one npx command) to your agent, then approve a browser consent screen — OAuth 2.1 with PKCE. You choose the organization and the scopes.

Ask your agent

Talk to it in plain language: “Draft a pipeline for a Senior Backend Engineer,” “who’s waiting on review?”, “invite the top three.” No new dashboard to learn.

It runs the pipeline

The agent calls real tools on your behalf — creating pipelines, sending real invitation emails, reading evaluations, and (only if you grant it) recording hiring decisions.

What your agent can do

37 tools span the full hiring workflow. Each group below lists representative tool names — the exact identifiers your agent calls.

Discover roles & rubrics

Explore role families, resolved stage sequences, and scoring dimensions before you build.

discover_role_familiesget_role_configlist_stage_typeslist_assessment_templates

Draft a JD & propose a pipeline

Turn a job description into a recommended, editable hiring pipeline — then create it.

analyze_jdgenerate_jdpropose_pipeline_from_jdcreate_pipelineadd_stageupdate_pipeline

Add & invite candidates

Add candidates, send (or re-send) stage invitations, and advance them through stages.

add_candidatesinvite_candidateadvance_candidate

Monitor live sessions

Track who's in progress, who needs attention, and how a pipeline's funnel is moving.

list_pipelineslist_candidatesget_candidateget_session_statusget_pipeline_activity

Read evaluations, reports & replays

Pull full evaluations, per-candidate reports, and signed session-replay manifests.

get_evaluationget_candidate_reportget_unified_reportget_replay_manifest

Make hiring decisions

Record hire / reject / manual-round decisions — gated behind a separate, opt-in scope.

hire_candidatereject_candidaterecord_manual_decision

Analytics, org & credits

See dashboards and leaderboards, and check your team, plan, and credit balance.

analytics_overviewleaderboardget_organizationget_credit_balancelist_team_members

Security & scopes

You connect through OAuth 2.1 with PKCE and approve a server-rendered consent screen bound to a single organization. Read-only by default; write actions are opt-in.

Granular OAuth scopes — read-only by default; you pick what to grant.
Server-rendered consent screen, bound to a single organization.
Hire / reject decisions sit behind a separate, opt-in scope.
Every query is scoped to your org; demo data is always excluded.
Server-side tokens stored hashed; local helper tokens stay on your machine (0600) and can be revoked.
Every action recorded in the event-sourced audit trail.

Frequently Asked Questions

What is the Model Context Protocol (MCP)?

MCP is an open standard that lets AI agents securely connect to external tools and data. InterviewLM exposes its hiring surface as a remote MCP server, so any MCP-capable agent — Claude Code, Cursor, Codex, Antigravity, GitHub Copilot, Claude Desktop, or ChatGPT — can run your hiring workflow through a set of well-defined, permissioned tools.

Which AI agents can I use with InterviewLM's MCP server?

Any MCP-capable client. We have one-line setup for Claude Code, Codex, GitHub Copilot, Cursor, Antigravity, Claude Desktop, and ChatGPT. Native-remote clients connect to the hosted endpoint directly and run OAuth in-browser; stdio-only clients use the npx @interviewlm/mcp helper, which handles OAuth for them.

Is it secure — can the agent act without my approval?

You connect through OAuth 2.1 with PKCE and approve a consent screen that lists exactly which scopes the agent gets, bound to one organization. Hire/reject decisions sit behind a separate opt-in scope. Every request is scoped to your org, server-side tokens are stored hashed, local helper tokens stay on your machine and can be revoked, and every action is recorded in the audit log.

Do I need to write code to use it?

No. There are no API keys or manual tokens to manage. You add the connector URL (or one npx command) to your agent and approve the consent screen — then you work in plain language.

What does it cost?

The MCP server is included with InterviewLM. You only pay for assessments you run — from $4.00 per AI interview, with 3 free credits to start and no credit card required. Reading pipelines, candidates, and reports through the MCP server doesn't cost credits; sending a candidate to a stage does, exactly as it would in the app.

Can the agent actually make hiring decisions?

Only if you grant the separate decision scope at consent time. With it, the agent can record hire, reject, and manual-round decisions — the same actions you'd take in the dashboard, written through the same services, and always attributed to you in the audit trail.

Back to Docs ATS Integrations

Overview

Remote endpoint

https://interviewlm.com/api/mcp

stdio helper

npx -y @interviewlm/mcp

npm: @interviewlm/mcp — for stdio-only clients.

Connect your agent

Claude Code

Terminal

Agent docs →

Native remote connector — Claude Code runs the OAuth consent for you.

Run in your terminal

claude mcp add --transport http interviewlm https://interviewlm.com/api/mcp

Codex

Terminal

Agent docs →

stdio via the @interviewlm/mcp helper — OAuth opens in your browser on first run.

Run in your terminal

codex mcp add interviewlm -- npx -y @interviewlm/mcp

GitHub Copilot

Terminal

Agent docs →

Native remote connector in VS Code — Copilot handles OAuth.

Run in your terminal (VS Code CLI)

code --add-mcp '{"name":"interviewlm","type":"http","url":"https://interviewlm.com/api/mcp"}'

Cursor

Config file

Agent docs →

Remote server — Cursor runs the OAuth consent in-browser.

Add to .cursor/mcp.json

{
  "mcpServers": {
    "interviewlm": {
      "url": "https://interviewlm.com/api/mcp"
    }
  }
}

Antigravity

Config file

stdio via the @interviewlm/mcp helper.

Settings → Customizations → Open MCP Config (mcp_config.json)

{
  "mcpServers": {
    "interviewlm": {
      "command": "npx",
      "args": ["-y", "@interviewlm/mcp"]
    }
  }
}

Claude Desktop

Connector

Add as a custom connector — works in Claude Desktop and claude.ai.

Settings → Connectors → Add custom connector

https://interviewlm.com/api/mcp

ChatGPT

Connector

Add as a custom connector (Plus, Pro, Business & Enterprise).

Settings → Connectors → Add custom connector (Developer mode)

https://interviewlm.com/api/mcp

How it works

Connect

Ask your agent

Talk to it in plain language: “Draft a pipeline for a Senior Backend Engineer,” “who’s waiting on review?”, “invite the top three.” No new dashboard to learn.

It runs the pipeline

The agent calls real tools on your behalf — creating pipelines, sending real invitation emails, reading evaluations, and (only if you grant it) recording hiring decisions.

What your agent can do

37 tools span the full hiring workflow. Each group below lists representative tool names — the exact identifiers your agent calls.

Discover roles & rubrics

Explore role families, resolved stage sequences, and scoring dimensions before you build.

discover_role_familiesget_role_configlist_stage_typeslist_assessment_templates

Draft a JD & propose a pipeline

Turn a job description into a recommended, editable hiring pipeline — then create it.

analyze_jdgenerate_jdpropose_pipeline_from_jdcreate_pipelineadd_stageupdate_pipeline

Add & invite candidates

Add candidates, send (or re-send) stage invitations, and advance them through stages.

add_candidatesinvite_candidateadvance_candidate

Monitor live sessions

Track who's in progress, who needs attention, and how a pipeline's funnel is moving.

list_pipelineslist_candidatesget_candidateget_session_statusget_pipeline_activity

Read evaluations, reports & replays

Pull full evaluations, per-candidate reports, and signed session-replay manifests.

get_evaluationget_candidate_reportget_unified_reportget_replay_manifest

Make hiring decisions

Record hire / reject / manual-round decisions — gated behind a separate, opt-in scope.

hire_candidatereject_candidaterecord_manual_decision

Analytics, org & credits

See dashboards and leaderboards, and check your team, plan, and credit balance.

analytics_overviewleaderboardget_organizationget_credit_balancelist_team_members

Security & scopes

You connect through OAuth 2.1 with PKCE and approve a server-rendered consent screen bound to a single organization. Read-only by default; write actions are opt-in.

Granular OAuth scopes — read-only by default; you pick what to grant.
Server-rendered consent screen, bound to a single organization.
Hire / reject decisions sit behind a separate, opt-in scope.
Every query is scoped to your org; demo data is always excluded.
Server-side tokens stored hashed; local helper tokens stay on your machine (0600) and can be revoked.
Every action recorded in the event-sourced audit trail.

Frequently Asked Questions

What is the Model Context Protocol (MCP)?

Which AI agents can I use with InterviewLM's MCP server?

Is it secure — can the agent act without my approval?

Do I need to write code to use it?

No. There are no API keys or manual tokens to manage. You add the connector URL (or one npx command) to your agent and approve the consent screen — then you work in plain language.

What does it cost?

Can the agent actually make hiring decisions?