AI Applyd MCP Server

AI Applyd MCP Server provides ATS resume scoring, interview prep, resume optimization, cover letters, job search, and auto-apply tools for AI assistants. Connect any MCP-compatible client (Claude, ChatGPT connectors, Cursor) to get AI-powered career tools directly in your workflow.

Public listing page: https://aiapplyd.com/mcps

Canonical Endpoints

Use /mcp (Streamable HTTP) for modern clients. Use /sse only for legacy clients that cannot speak Streamable HTTP - typically via the mcp-remote bridge.

Related MCP server: workopia-mcp

Tools, Tiers & Funnel

There is no free tier and no anonymous tool. Every tool requires a connected AI Applyd account, and every tool is metered against that user's own plan and paid for with their own AI credits. The API enforces this on every call (requireActionAllowed → requireAIPrerequisites → incrementActionUsage / reserveUserTokens); the MCP server holds no allowance of its own.

Also exposed: 3 prompts (review_my_resume, prepare_for_interview, find_jobs_like_this) and 3 resources (ats-best-practices, interview-frameworks, resume-section-order).

The funnel:

1. Connect - sign in with Google once (a free AI Applyd account, no card). New accounts receive a one-time 10,000-token starter grant, which is what the first tool calls draw from.

2. Use - every tool spends that balance, exactly as the dashboard does.

3. Upgrade - when credits run low, tools return an upsell to continue in the dashboard at https://aiapplyd.com (carrying per-client UTM attribution, so we can see which assistant drives signups).

Two behaviours worth knowing

• aiapplyd_search_jobs is read-only. It never rewrites your saved preferences. To change what AI Applyd hunts for, call aiapplyd_update_job_preferences — which is annotated destructiveHint: true because a supplied list replaces the stored one.

• aiapplyd_auto_apply applies to the one job URL you give it, and follows the review setting already on your account (auto-approve submits on its own; copilot routes it to your review queue). It never changes that setting.

Connect from Claude (web + Desktop + Code)

AI Applyd is a remote MCP server. The modern path is the built-in custom connector with OAuth - no local bridge needed.

Claude web / Claude Desktop (custom connector)

1. Open Settings → Connectors → Add custom connector.

2. Name: AI Applyd.

3. Remote MCP server URL: https://mcp.aiapplyd.com/mcp

4. Save. The first tool call opens the OAuth flow - sign in with Google (a free account is enough for the no-credit tools) and the token is picked up automatically.

If your Claude build does not yet support Streamable HTTP custom connectors, use the legacy mcp-remote bridge in claude_desktop_config.json:

{
  "mcpServers": {
    "ai-applyd": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.aiapplyd.com/sse"]
    }
  }
}

Claude Code

Add the remote server (Streamable HTTP, recommended):

claude mcp add --transport http ai-applyd https://mcp.aiapplyd.com/mcp

Or via .claude/settings.json using the legacy bridge:

{
  "mcpServers": {
    "ai-applyd": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.aiapplyd.com/sse"]
    }
  }
}

Connect from ChatGPT

ChatGPT has two distinct integration paths. Pick based on what you have access to:

Native MCP connector (ChatGPT connectors)

Newer ChatGPT connectors consume the MCP protocol directly. Add a custom connector pointing at:

https://mcp.aiapplyd.com/mcp

OAuth is handled in-app - sign in with Google to unlock connected and paid tools.

Custom GPT via Actions (OpenAPI)

Not supported. ChatGPT Custom GPTs consume REST through Actions, and this server intentionally exposes no anonymous REST surface (every capability requires the OAuth-authenticated MCP protocol). Use the native MCP connector path above instead.

Connect from Cursor

Go to Settings → MCP → Add new MCP server and add a remote server:

• URL (modern): https://mcp.aiapplyd.com/mcp

For Cursor builds without Streamable HTTP support, use the legacy bridge:

{
  "mcpServers": {
    "ai-applyd": {
      "command": "npx",
      "args": ["mcp-remote", "https://mcp.aiapplyd.com/sse"]
    }
  }
}

Any MCP Client

• Streamable HTTP (modern): https://mcp.aiapplyd.com/mcp

• SSE (legacy): https://mcp.aiapplyd.com/sse

• Server metadata / registry card: https://mcp.aiapplyd.com/.well-known/mcp/server-card.json

Authentication

Every tool call requires a connected AI Applyd account. There are no anonymous tools and no free compute: an anonymous tools/call is answered with HTTP 401 + WWW-Authenticate (which is what makes hosted clients offer "Connect"), while discovery (initialize, tools/list) stays open so a connector can read the catalog and bootstrap OAuth.

You are prompted to sign in with Google the first time you invoke a tool. AI Applyd implements OAuth 2.1 with Dynamic Client Registration:

• Discovery: /.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource

• Registration (DCR): POST /register

• Authorize: GET /authorize — PKCE S256 is mandatory; there is no plain and no non-PKCE path

• Token: POST /oauth/token — authorization_code and refresh_token grants

• Revoke: POST /oauth/revoke (RFC 7009)

Security properties, all covered by tests:

• Short-lived access tokens (1h) plus a rotating refresh token (90d). Each refresh issues a new refresh token and consumes the old one.

• Refresh-token reuse detection. Replaying a consumed refresh token means it leaked, so the entire token family is burned and the client must re-authorize.

• Tokens are hashed at rest. KV stores SHA-256(token), never the token, so a KV dump yields no usable credential.

• Revocation kills the credential when a user disconnects the connector, instead of leaving it live until expiry.

• Redirect allowlist on both /authorize and DCR, so an open registration endpoint cannot smuggle in an attacker-owned redirect.

• Rate limits on /register, /authorize, /oauth/token and /oauth/revoke (all unauthenticated, all KV-writing).

Paid tools additionally require an active AI Applyd subscription with available token balance.

Rate Limits

• Tools are governed by your AI Applyd plan: per-action allowances and your token balance, enforced API-side. See pricing.

• The unauthenticated OAuth surface is rate limited per IP. If you hit a limit, back off and retry.

Status

This is a hosted remote MCP server. There is nothing to install and no local build step: point any MCP-compatible client at https://mcp.aiapplyd.com/mcp and complete the OAuth flow.

This repository is the public home for the server's manifest (server.json), its documentation and its connection instructions. The implementation runs as a Cloudflare Worker operated by AI Applyd.

Links

• AI Applyd

• MCP listing page

• Pricing

• Documentation