Engine
Measure bias, stability, and consistency of model outputs and workflows through 68 MCP tools spanning finance, gaming, AI, and crypto domains.
Overview
ZPL Engine is a Ship-phase MCP server that offers 68 bias-and-stability tools across 11 categories for reviewing AI and agent outputs before release.
What is this MCP server?
- 68 tools across 11 categories for AIN bias and stability (finance, gaming, AI, crypto)
- npm zpl-engine-mcp v4.2.1 with ZPL_API_KEY (zpl_u_ user keys; service keys rejected since v3.5.0)
- ZPL_MODE pure (default) hides AIN score on text-evaluation tools to reduce observer effect; coach exposes score
- Eight AI-eval tools may require ANTHROPIC_API_KEY with a 100 Claude calls per process session cap
- Default engine base URL https://engine.zeropointlogic.io overridable via ZPL_ENGINE_URL
- Package version 4.2.1
- 8 AI-eval tools may use Anthropic with 100 calls per process cap
What problem does it solve?
You cannot manually spot instability, sycophancy, or domain bias across every agent response your solo product generates.
Who is it for?
Builders shipping LLM features in regulated-flavored or high-stakes domains who want quantitative bias/stability passes in the MCP loop.
Skip if: Simple CRUD apps with no generative AI, or teams without budget for ZPL API keys and optional Claude eval usage.
What do I get? / Deliverables
Your agent can invoke categorized ZPL evaluations—with pure or coach scoring modes—using a keyed connection to the ZPL engine API.
- Bias and stability scores across 11 tool categories
- Configurable pure vs coach evaluation behavior
- Repeatable pre-ship AI output review via MCP
Recommended MCP Servers
Journey fit
Ship review is where you judge whether agent-generated content and decisions are stable before customers see them; evaluation belongs at the gate. Review covers systematic checks on AI behavior—not unit tests alone but bias/stability scoring across categorized engine tools.
How it compares
AIN evaluation MCP with 68 specialized tools, not a linter, unit-test framework, or hosting panel.
Common Questions / FAQ
Who is ZPL Engine for?
Solo developers and small teams running agent or LLM features who need structured bias and stability measurements via MCP rather than ad-hoc prompt vibes.
When should I use ZPL Engine?
Use it during ship review and post-change iteration when you must score consistency, sycophancy, or domain stability before users rely on generated outputs.
How do I add ZPL Engine to my agent?
Create a zpl_u_ user API key in the Zero Point Logic dashboard, set ZPL_API_KEY and optional ZPL_MODE/ZPL_ENGINE_URL, install zpl-engine-mcp, add ANTHROPIC_API_KEY only if you use the eight Claude-backed eval tools, then register stdio MCP in your host.