Schemas teach the moves. Pilotage teaches the game.
CoolRIOTS releases Pilotage, an open standard that teaches AI agents to use complex tools
CoolRIOTS today published Pilotage, an open extension of the Model Context Protocol (MCP). Where a schema only tells an AI agent what a tool takes, Pilotage teaches it how to actually use one — with guides to learn from, side-effect-free validation, planning with risk flags, deliberate execution, and a traceable audit record. The specification and a reference implementation are available now.
CoolRIOTS today announced the release of Pilotage, an open standard that extends the Model Context Protocol (MCP) so AI agents can learn, validate, plan, execute, and trace their use of complex tools through a single, consistent interface. The specification and a reference implementation are publicly available from today.
MCP is how AI agents plug into real systems and are handed the tools to act. But a tool's schema only describes structure — which fields exist and what type they are. It cannot teach the grammar, the valid names, or the rules a real system enforces, and it gives an agent a single blind attempt with no way to draft, check, or repair its work before something happens for real. Pilotage closes that gap by letting the server that owns a tool teach how to use it — once, for every agent — instead of every agent guessing on its own.
Schemas teach the moves. Pilotage teaches the game. Giving an agent a tool is not the same as teaching it to use one well — and on real enterprise systems, the difference is everything. Pilotage lets the server do the teaching, so a Digital Employee can check its work before it acts, and leave a record of exactly what it did. We published it as an open standard because trustworthy AI shouldn't be a proprietary advantage.
Dr Sharon Lim, CEO of CoolRIOTS
A competence layer for MCP
Most real work is not a single value typed into a box — it is a small program: a query, a workflow, an automation rule, a filing. Today, the competence to build those programs is rebuilt inside every agent through custom prompts and hand-built integrations, an N×M cost that repeats for every agent and every system. Pilotage moves that knowledge to where it belongs: the MCP server, which already owns it.
Pilotage does this through seven small capabilities a server can offer — a hub the agent drives, not a pipeline it must follow:
- Manifest — the handshake: the server declares what it can teach and how.
- Guides — navigable documentation the server publishes about its own tools.
- Catalog — a live, versioned view of the real environment: the names and options that exist right now.
- Validate — the agent checks a draft with zero side effects and gets clear diagnostic codes before anything runs.
- Plan — the server returns the exact steps it would take, each annotated with its risk, so destructive actions are seen and approved before they happen.
- Execute — the real work runs, only once the draft has been checked and the plan understood.
- Trace — every run leaves a verifiable record, stored by run ID: what was decided, why, and what happened.
Additive to MCP, honest about trust
Pilotage is entirely additive. It introduces no new transport and no new session model, and is negotiated through MCP's own extension mechanism — if a system already speaks MCP, Pilotage can be adopted without rearchitecting it. It standardises the loop, not the language: an organisation's SQL stays SQL and its workflow formats stay its own. The specification is also deliberately honest about its limits — Pilotage makes a cooperative server more understandable and more checkable; it does not claim to make a malicious server safe.
Why CoolRIOTS built it
A Digital Employee is only as trustworthy as the way it uses an organisation's systems. Validate-before-execute, risk-flagged plans, and traces stored by run ID make that use safer and auditable by default — reinforcing the Governance and Openness commitments at the heart of the CoolRIOTS GOOD Framework. By publishing Pilotage as an open standard rather than a proprietary feature, CoolRIOTS invites the wider ecosystem — any MCP server, not only its own — to build on it.
Availability
The Pilotage specification is open and available now, alongside a reference implementation and a public Harborview example server that lets developers try Pilotage workflows against a live, simulated environment.
Explore Pilotage as a CoolRIOTS capability
Read the Pilotage specification
About CoolRIOTS
CoolRIOTS delivers agentic AI workflow orchestration and knowledge management for the enterprise through its BeX Enterprise AI platform, turning fragmented workflows into a governed Digital Workforce — with outcomes delivered in 45 days.
Media enquiries: [email protected]