Agent-native authoring over MCP: Gessa vs Unity vs Unreal
Last verified 2026-09-06 against engine v1.0.232. Competitor sources were accessed 2026-09-06; every Gessa cell binds to the generated engine reference and the build fails if a bound surface disappears.
Gessa is authored over the Model Context Protocol: an agent calls the same capability surface a human does, through the same action catalog and the same semantic-patch layer, and the two are held to a parity floor by the build. Neither Unity nor Unreal ships a first-party MCP or agent-native authoring surface; both author through a human-facing editor and scripting system (C# and Visual Scripting in Unity, Blueprints and C++ in Unreal). The honest tradeoff is that Gessa's physics-family authoring depth is thinner than either competitor, and agent authoring is bounded to the published capability surface rather than arbitrary engine-source edits.
The MCP tool surface is in the MCP tools reference; the action verbs are in the action catalog reference, and behavior authoring is explained in Script IR and GessaScript.
Where Gessa stands
| Capability | Gessa | Unity | Unreal |
|---|---|---|---|
| Author over the Model Context Protocol | Supportedcheck:mcp-coverage | No first-party MCP authoring unity 6000.3 | No first-party MCP authoring unreal 5.8 |
| Human and agent share one capability surface | Supportedcheck:ai-mcp-sdk-authoring-parity | Human scripting only unity 6000.3 | Human scripting only unreal 5.8 |
| Component authoring with the same verbs a human uses | Supportedcheck:component-verb-ai-parity | Editor and scripting unity 6000.3 | Editor and Blueprints unreal 5.8 |
| Behavior authoring via semantic patch over a canonical IR | Supportedcheck:script-semantic-patch-coverage | Visual scripting and C# unity 6000.3 | Blueprints visual scripting unreal 5.8 |
| World build from a typed spec | Supportedcheck:world-build-contract-catalog | Editor authoring unity 6000.3 | Editor authoring unreal 5.8 |
| One action catalog for human, agent, MCP, and SDK | Supportedcheck:action-catalog-coverage | No unified agent catalog unity 6000.3 | No unified agent catalog unreal 5.8 |
| Physics-family authoring depth | Partialcomponent.joint.contract.v1 | Supported unity 6000.3 | Supported unreal 5.8 |
The MCP, parity, component-verb, semantic-patch, world-build, and action-catalog rows are backed by the check:mcp-coverage, check:ai-mcp-sdk-authoring-parity, check:component-verb-ai-parity, check:script-semantic-patch-coverage, check:world-build-contract-catalog, and check:action-catalog-coverage gates in the build; the physics-depth row binds to the JointComponent contract, which is itself partial.
The gaps, named here
- Physics-family authoring depth is thinner than either competitor. The
JointComponentcontract carries a limited set of joints and no articulation body, so the physics-depth row reads partial. The full physics comparison, including the families Gessa does not simulate, is on the deterministic physics page. - Agent authoring is bounded to the capability surface. An agent authors through the same published action catalog and semantic-patch layer a human uses; it does not edit engine source or reach outside that surface. The parity floor is a floor for the authoring surface, not a claim that an agent can do anything a native engine programmer can do.
- Parity is proven for authoring, not for every runtime subsystem. The parity gates cover the authoring capabilities; they do not assert that every downstream renderer or physics family exists, which is why this page names the physics-depth concession on the same page.
Where Unity and Unreal are ahead
- Unreal's Blueprints visual scripting and C++ give designers and programmers a deep, mature, human-facing authoring system with a very large surface. See the Unreal Blueprints documentation cited in the table.
- Unity's C# scripting and Visual Scripting are a mature human-facing authoring path with a large ecosystem. See the Unity scripting documentation cited in the table.
- Both engines carry far broader physics, rendering, and animation family catalogs than Gessa, so a human author there can reach depth Gessa does not offer.
Neither competitor, though, exposes a first-party MCP or agent-native authoring surface with a human-and-agent parity floor, which is the wedge this page leads with. The other comparison axes are in the compare index.