For AI agents and integrators

Psychometric assessments for your AI agents, over MCP

Plug your AI agent into Talent Assessment with the Model Context Protocol. Create campaigns, invite candidates, fetch structured results. API-key auth, pay-per-assessment billing.

Read the docs →
MCP 2025-03JSON-RPC 2.0GDPR-readyBig Five · Schwartz · OCP
AI agent connected to Talent Assessment via MCP

Why MCP, and why here

🤖

Built for agents

Typed tools, clear descriptions, JSON schemas. No scraping, no web sessions, no glue code.

🧪

Real psychometrics

Big Five, Schwartz, OCP, documented predictive validity. Your agents don't ship a BuzzFeed quiz.

💳

Clean billing

One key per project, usage counter, Stripe metered. Resell to your clients without plumbing.

Exposed MCP tools

Every tool is typed, documented, and callable from Claude, Cursor, ChatGPT, or any MCP client.

screening.create_campaign

Create a candidate screening campaign for a given job, returns a campaign id and shareable invite URL.

screening.invite_candidate

Send an email invitation to a candidate for a given campaign.

screening.get_results

Fetch structured results (Big Five scores, fit score, summary) for a completed candidate.

profiling.create_campaign

Create an organizational profiling campaign to map team values and culture.

profiling.get_org_profile

Return the aggregate culture profile (Schwartz, OCP, Quinn) for an organization.

frameworks.list

List all available frameworks and their score dimensions.

60-second setup

  1. Get an API key. Sign up free and mint a key from your dashboard.
  2. Add the server to your MCP client.
    {
  "mcpServers": {
    "talent-assessment": {
      "url": "https://api.talentassessment.io/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}
  3. Call the tools. Ask your agent: "Spin up a Product Manager campaign and invite [email protected]".

Example: a recruiting agent 

// Agent receives: "Run screening for Alice on the PM role."

const campaign = await call('screening.create_campaign', {
  job_slug: 'product-manager',
  frameworks: ['big-five', 'schwartz'],
})

await call('screening.invite_candidate', {
  campaign_id: campaign.id,
  email: '[email protected]',
  candidate_name: 'Alice Martin',
})

// Later, after Alice completes the test:
const results = await call('screening.get_results', {
  campaign_id: campaign.id,
  candidate_email: '[email protected]',
})

// => { fit_score: 0.82, traits: {...}, summary: "..." }

A human respondent stays in the loop: the agent orchestrates, the candidate takes the test.

Agent shortlisting candidates

API pricing

No subscription, no commitment. You pay for completed assessments.

Starter

€0/ first 10 assessments

For trying the integration.

  • 10 free assessments
  • All MCP tools
  • Community support

Pay as you go

€9/ completed assessment

The default for most integrations.

  • Stripe-metered billing
  • No completion = no charge
  • Email support
  • Completion webhooks

Volume

Talk to us

Above 500 assessments / month.

  • Volume pricing
  • Contractual SLA
  • Optional white-label

Stripe-metered, billed on completion. Invitations sent but not finished are not billed.

Growth and usage-based billing

FAQ

What is MCP?

Model Context Protocol is an open standard for connecting AI agents to external tools, designed by Anthropic. Any MCP-aware client (Claude, Cursor, ChatGPT desktop, custom agents) can talk to our server with one config block.

Can an agent take the test for the candidate?

No, and it shouldn't. The respondent is always a human. The agent creates campaigns, sends invitations, and reads results once the candidate has completed the questionnaire. The assessment retains its predictive validity.

How is billing handled?

Pay-per-completion via Stripe metered. Each API key is tied to a project and emits a usage event when a candidate finishes an assessment. No completion = no charge.

Is it GDPR-friendly?

Yes. Candidate data is stored in the EU (Hetzner Falkenstein), results can be deleted on request, and we provide a DPA. The MCP server only exposes endpoints that already pass our existing access controls.

Which clients work?

Any MCP-compliant client: Claude Desktop, Claude Code, Cursor, Windsurf, ChatGPT desktop (with MCP enabled), and custom agents built on the MCP SDK in TypeScript or Python.

Where can I see the source?

The MCP server is a thin layer over our public REST API. The OpenAPI spec is published on the dashboard once you have a key. The MCP wrapper itself will be open-sourced.

Wire your agent in 5 minutes

Free account, instant API key, 10 free assessments to try.

No credit card required