// comparison
Short version: they solve overlapping but different problems, and they are usually complementary. Orchestrators like Temporal, Restate and LangGraph implement compensation inside your own workflow code. ChronoMCP is a transparent guard at the protocol level (MCP) — it works with any agent and any tool, adds human approval and an honest view of what cannot be undone, without you rewriting anything.
A saga orchestrator is a framework you build your workflow on top of: you write the steps and their compensations in code, and the engine runs them durably. ChronoMCP is a proxy that sits between an AI agent and its tools (over the Model Context Protocol) and governs each tool call — previewing impact, gating risky mutations on a human, rolling back declared-reversible steps on failure, and telling the truth about the irreversible ones. You don't build your agent "on" ChronoMCP; you put it in front of whatever agent you already have.
| ChronoMCP | Orchestrators (Temporal, Restate) |
Agent frameworks (LangGraph) |
DIY / native MCP annotations |
|
|---|---|---|---|---|
| Where compensation lives | Protocol-level proxy (transparent) | Inside your workflow code | Inside your agent graph | Ad-hoc in your code |
| Works with any MCP server unchanged | Yes | N/A (not MCP-specific) | Depends on your graph | Manual, per tool |
| Human approval gate + impact diff | Built-in (terminal, policy, Slack quorum) | No (you build it) | No | No |
| Honesty about irreversibility | Declared + flagged before commit | Not modeled | Not modeled | Manual |
| LLM in the rollback path | Never — deterministic (anti prompt-injection) | Your code decides | Often the LLM decides | Depends |
| Tamper-evident audit | Hash-chained, anyone verifies | Workflow history (trust the server) | App logs | Manual |
| Setup cost | npm i -g chronomcp, transparent | Significant (SDK, workers, infra) | Adopt the framework | Per-tool code |
| Best for | Guarding any agent's tool calls with human oversight | Complex, long-running business workflows you own | Building the agent's reasoning graph | Small, fully-controlled cases |
green = strength · amber = partial/depends · red = not designed for it. This compares categories; every tool listed is excellent at what it was built for.
The honest verdict: ChronoMCP is the guardrail at the boundary between an AI agent and the real world; an orchestrator is the engine that runs a workflow you built. Many teams run both — the orchestrator drives the durable business process, and ChronoMCP guards the individual, risky tool calls the agent makes, with a human able to say "no" before anything irreversible happens.
Most "undo" stories quietly break at the edges: a refund is a new transaction, not a reversal; a sent email cannot be un-sent. ChronoMCP makes irreversibility a first-class, declared property (via the open mcp-compensate extension) and surfaces it before you approve — not as a surprise after a failure. That is the difference between a guard you can trust and a rollback that lies.