UluOps Documentation
TypeScript SDKs, MCP clients, and CLI tools for the UluOps validation platform. Encode expert judgment. Track quality over time. Improve continuously.
UluOps is in active development — APIs are stabilizing but may still change between minor versions for pre-1.0 packages.
First validation run
Get from zero to working results in 5 minutes.
Get started →Using Claude Code?
Connect via MCP — 92 tools, zero config beyond API key.
MCP docs →Integrate via SDK
TypeScript SDKs for registry and validation tracking.
SDK docs →REST API reference
Direct HTTP access to Platform and Registry APIs.
API docs →SDK Packages
Four packages covering the full UluOps surface area.
@uluops/core
The execution engine — run agents, commands, workflows, and pipelines against your code.
0.24.3 →@uluops/registry-sdk
Manage definitions — agents, commands, workflows, pipelines.
0.36.0 →@uluops/ops-sdk
Track validation runs, manage issues, analyze trends. 70 methods.
5.0.0 →@uluops/sdk-core
HTTP client, auth, and error infrastructure. For SDK authors.
0.13.0 →MCP Clients
For AI agents and Claude Code — thin protocol clients backed by the same APIs as the SDKs.
uluops-ops-mcp
48 tools for validation tracking. Save runs, query issues, analytics, analysis records. Zero business logic — thin client over ops-sdk backend.
48 tools · 2 resources →uluops-registry-mcp
MCP client for the Registry API. Browse, fetch, publish definitions, and track effectiveness from Claude Code.
44 tools →Platform Architecture
Two parallel access lanes to the same backend APIs:
Key Concepts
| Concept | Description |
|---|---|
| Definitions | Versioned YAML specs for agents, commands, workflows, and pipelines (ADL/CDL/WDL/PDL) |
| Failure Taxonomy | Four-domain classification: Structural, Semantic, Pragmatic, Epistemic |
| Fingerprinting | SHA-256 hash correlation tracking issues across runs as new, recurring, or regression |
| Recursive Appreciation | How improvements to agents cascade through the ecosystem |
New to UluOps? Start with the Quickstart — it walks through your first validation run end-to-end.