Integrations — What Cromus Works With
Cromus integrates with your stack in four ways: it compiles governed skills to agent platforms, connects to assistants through a hosted MCP server, reasons about model providers through a verified registry, and exports workflows to open-source frameworks. Every capability listed is shipped.
Skill Studio's Platform Selector compiles a single Cromus SKILL.md to six targets, each with platform-specific validation: Cromus Native, Replit Agent (the .agents/skills/ layout), AgentSkills.io, OpenClaw (ClawHub registry), Claude Managed Agents (Anthropic Agent endpoint config), and Google AI Studio (Gemini API config plus the Interactions API typed-steps template). The governance travels with the artifact; Cromus never sits between your agent and the model.
Cromus runs a hosted, streamable MCP server with 15 deterministic tools and zero LLM inference calls — usable from Claude, Claude Cowork, Claude Code, and OpenAI Codex CLI, or any MCP client via OAuth 2.1 (PKCE-mandatory) or a legacy API key with read/score/write scopes.
Cromus reasons about cost across a verified model registry spanning major providers, grouped into four ecosystem tiers by setup effort: Mainstream, Managed, Aggregator, and Infra. Provider Scope ensures estimates and recommendations respect the providers a workspace permits. The Graph Layer normalizes nine import formats into Graph IR for bounded path simulation, path budgets, and Reality Check calibration after validated observations. The Open Workflow Ecosystem exports to LangGraph, CrewAI, AutoGen, n8n, and more. The deepest integration is the open-spec contract itself: SKILL.md, ETHOS.md, MEMORY.md, and GRAPH.md, each with a free public validator. Cromus is upstream of the runtime — it compiles, scores, and governs before execution, and never proxies model inference calls.