Labsco
MCP SERVER

ZPL Engine MCP

by cicicalex ↗

Score text, datasets, portfolios or model responses for balance and stability, and get the number back rather than an opinion.

LLM Evaluation & Observability
Summary
The default mode hides the score from the model that is being scored.

That is the design decision worth knowing: in pure mode the text-evaluation tools withhold the number from the assistant so its next response is not shaped by it. The project also removed five tools in an earlier release because they read as predictions or certifications — the score is a stability measurement, not a recommendation.

What it is

A client for the Zero Point Logic Engine, a stability and neutrality analysis service. Tools across eleven categories send content or numbers to the engine and return a score with the supporting breakdown.

What you get
  • 69 tools across 11 categories: 65 unique names plus 4 backwards-compatible aliases
  • Core: zpl_compute, zpl_sweep, zpl_matrix, zpl_analyze, zpl_domains, zpl_health, zpl_history, zpl_watchlist, zpl_report
  • Finance: zpl_market_scan, zpl_portfolio, zpl_fear_greed, zpl_forex_pair, zpl_sector_bias, zpl_macro, zpl_correlation
  • Gaming: zpl_loot_table, zpl_matchmaking, zpl_economy_check, zpl_pvp_balance, zpl_gacha_audit, zpl_rng_test
  • AI/ML and Security: zpl_model_bias, zpl_dataset_audit, zpl_prompt_test, zpl_benchmark, zpl_vuln_map, zpl_risk_score, zpl_compliance
  • Crypto: zpl_whale_check, zpl_defi_risk, zpl_liquidity, zpl_tokenomics
  • Universal: zpl_check_response, zpl_explain, zpl_decide, zpl_compare, zpl_rank
  • AI Eval: zpl_consistency_test, zpl_sycophancy_score, zpl_refusal_balance, zpl_language_equity, zpl_persona_drift, zpl_safety_boundary, zpl_hallucination_consistency, zpl_emotional_stability
  • Meta: zpl_about and zpl_validate_input work without auth and cost nothing; zpl_quota, zpl_score_only, zpl_diagnose, zpl_batch, zpl_export, zpl_usage, zpl_account
Requirements

Npx and the package zpl-engine-mcp over stdio. ZPL_API_KEY is required — the bundled setup command opens a browser, creates the key and patches the config of each installed client. ZPL_MODE defaults to pure, which hides the score from the model on text-evaluation tools; coach exposes it. The eight AI-eval tools additionally need ANTHROPIC_API_KEY and are capped per process.

Setup effort

One command plus a key — npx zpl-engine-mcp@latest setup, then supply credentials