Developers
Build on the game platform, not around it.
Fusionery exposes the same game-authoring tools its own planner uses, so AI agents, internal tools, and production systems can create, inspect, revise, build, and publish games through a bounded, metered contract.
MCP in beta
A deliberate surface for integration.
The Model Context Protocol endpoint is live for every workspace. Every surface here is built around bounded actions, scoped access, auditable changes, and the same plan limits and release-safety path used inside Fusionery Studio.
MCP · Beta
Connect AI clients to game work.
Point Claude, Cursor, or your own agent at /mcp with a workspace API token and it can list your games, inspect a game's summary and diagnostics, create a game from a concept, revise it with natural language or precise tool calls, queue a browser preview, and publish. Tokens are created by workspace owners and admins under Integrations, carry read, author, and platform scopes, and see only the projects their issuer can.
Live: JSON-RPC 2.0 over HTTP, 120+ scoped tools, structured errors for quota and missing capabilities, 120 requests per minute per token.
API · Via MCP today
Automate the production loop.
The MCP endpoint is also the programmatic API: each tool is a documented JSON-RPC call, so build pipelines and internal tools can drive creation, revision, builds, and releases without treating Fusionery as an opaque export step. A REST convenience layer is planned on top of the same contract.
Metered like the Studio: authoring calls count as AI revisions, builds as build runs, publishing against your plan's public-game limit. get_usage reports what remains.
Documentation · Beta
Ship against a clear contract.
The tool contract is generated from the platform itself: call tools/list (or the list_tools tool) for every tool's schema, owning capability, idempotency key, and whether it mutates or meters. Legacy tool names stay accepted through a published alias table, so integrations don't break when tools are renamed.
Available: the MCP contract and error codes. Planned: quickstarts, hosted-client OAuth, and a REST reference.
Beyond desktop control
Don’t ask an agent to improvise a production pipeline.
Desktop agents are excellent collaborators for bespoke local work: they can move between files, tools, and applications. Fusionery is better when the result needs to become a coherent, playable game with a durable source of truth—not just a set of changed files.
Raw desktop control
Flexible, but you own the glue.
- Directly operates the local editor, files, plugins, and command line.
- Depends on your project conventions, installed tools, and manual review.
- Can make a useful local change without proving the game still builds, plays, or releases safely.
Fusionery
A managed contract for making games.
- Agents operate bounded game concepts—systems, worlds, assets, builds, and releases—not arbitrary pixels and files.
- Every revision becomes a versioned, inspectable blueprint with capability checks, history, and undo.
- Preview builds, release readiness, workspace access, metering, and cross-platform delivery stay in the same production path.
The strongest workflow can use both: connect Claude Desktop or another MCP client to Fusionery, then let the agent use a game-aware platform surface instead of reconstructing your workflow from a local desktop. Claude Desktop supports MCP connections.
How creators earn
Your game. Your revenue.
Export your Fusionery-built game and sell it on Steam today—Fusionery takes no cut of your sales. Next on the roadmap: paid releases players can buy directly on Fusion Castle, with the majority of every sale going to the creator.
See how creators earn
Beta, with the same guardrails
The contract is published; the guardrails are the Studio's.
Agents act as the person who issued their token, inside that person's workspace and project access, against the same plan allowances. Every action is recorded in the game's history and can be undone from the Studio. The MCP surface is in beta: tool schemas are versioned through aliases, and we will announce breaking changes before making them.