Skip to main content
MCP (Model Context Protocol) lets an AI coding agent use Assess tools — create assessments, invite candidates, read results — with your permission. For docs search + skills + one-click install across agents, see Build with agents (Knowledge MCP is separate from this API server).

If you are not an engineer

You typically do not need MCP. Use the dashboard or ask an engineer to set up n8n / Automations. MCP is for people already working in Cursor, Claude Code, Claude Desktop, or Codex who want the agent to operate Assess safely.

Two ways to connect

The stdio package is @praxicraft/assess-mcp on npm. Hosted HTTP still runs in the bureau namespace at https://assess.praxicraft.com/mcp (image built from the web monorepo deploy mirror). Prefer least-privilege scopes on stored keys. IP allowlists on API keys apply to Public API ct_live_… / ct_test_… requests only. Hosted MCP OAuth is not IP-restricted. If you use stdio MCP with PRAXICRAFT_API_KEY, leave that key’s CIDR list empty (or use a dedicated MCP key).

Cursor / Codex

Claude Code

Claude Desktop (stdio)

What agents can do

Tools map to the Public API: assessments, invites, results, org/squads/audit (audit is Starter+), cases, webhooks, interviews, pipelines, integrations. Remind agents to respect plan limits and to branch on error.codeErrors.

Next

Build with agents · Using docs with LLMs · Quickstart · API Reference
Last modified on August 24, 2026