Mcp Server
Check agent or model outputs for fabrication and hallucination before you ship customer-facing AI features.
Overview
Arkheia MCP is a Ship-phase MCP server that flags fabrication and hallucination in LLM outputs across 30+ supported models.
What is this MCP server?
- Arkheia API-backed detection of fabrication and hallucination in LLM output
- Advertised support for 30+ models
- NPM package @arkheia/mcp-server v0.1.4 with stdio transport
- Requires ARKHEIA_API_KEY from arkheia.ai
- Server version 0.1.4
- 30+ models supported (per description)
- Transport: stdio
What problem does it solve?
Shipping agent features without a systematic hallucination check leaves solo founders one bad answer away from churn or compliance pain.
Who is it for?
Indie agent builders who need a quick external fidelity signal on model output across many LLM backends.
Skip if: Teams with no LLM-generated user content, or those who already run full offline eval harnesses with legal review baked in.
What do I get? / Deliverables
With ARKHEIA_API_KEY configured, your agent can run Arkheia checks on drafts and tighten prompts or retrieval before release.
- Structured hallucination or fabrication signals on candidate LLM outputs
- Higher-confidence ship decisions for customer-facing AI text
Recommended MCP Servers
Journey fit
How it compares
Third-party hallucination API via MCP, not an in-repo unit-test or prompt-engineering skill alone.
Common Questions / FAQ
Who is Arkheia MCP for?
Solo developers and small teams shipping LLM-powered products who want MCP-triggered hallucination checks during review.
When should I use Arkheia MCP?
Use it in Ship review when you finalize agent responses, marketing copy generated by models, or support automation before customers see it.
How do I add Arkheia MCP to my agent?
Get an API key at arkheia.ai, set ARKHEIA_API_KEY, install npm package @arkheia/mcp-server, and add the stdio server to Claude Code, Cursor, or Codex MCP config.