Tooling

Ketoy MCP

A knowledge layer for AI coding agents. It tells your agent what Ketoy supports before it writes a line.

The Ketoy MCP is a knowledge layer for AI coding agents. It answers one question with authority: is this supported in Ketoy, and if not, how do you bridge it. Your agent stops guessing and writes code that compiles to KBC.

It is read-only. It never writes code for you. It tells your agent what works, by what mechanism, and where the escape hatch is. The support matrix is extracted from the SDK source, so the answers stay accurate to the code.

Connect

Ketoy MCP is a standard streamable-HTTP MCP server. Any MCP-compatible AI coding agent can use it. The server is live at:

http://mcp.ketoy.dev/mcp

Add it to whichever agent you use — the URL is the same everywhere:

CLI agents: register it as an HTTP MCP server. For example, in Claude Code: claude mcp add --transport http ketoy http://mcp.ketoy.dev/mcp. Other CLI agents use their own mcp add / config equivalent.
Desktop and web agents: add it as a Custom Connector (HTTP) pointing at the URL above.
Editors and IDE agents (Cursor, Windsurf, Codex, and others): add an HTTP MCP server entry with the same URL.

If your tool speaks MCP over HTTP, it works — no plugin, no vendor lock-in.

What it needs

Nothing. No account, no API key, no database. It is pure compute serving bundled knowledge, hosted on Cloudflare Workers. Supported MCP protocol versions: 2025-06-18, 2025-03-26, 2024-11-05.

Which agent should I use?

Whichever one you already work in. Ketoy MCP is agent-neutral and we do not endorse any model or vendor. For what it is worth, when we exercised the tools end to end we saw the strongest results with Claude Code on the Opus 4.8 model — but that is an observation from our own testing, not a recommendation. Use the agent and model that fit your workflow.

Tools

Point your agent at these. ketoy_overview and ketoy_check_support carry most of the value.

Tool	Use it for
`ketoy_overview`	Orientation and the adapter, capability, and language decision tree. Call it first.
`ketoy_check_support`	Ask before using any Compose component or API. Supported? by what mechanism? escape hatch if not.
`ketoy_validate`	Paste a snippet for a heuristic pre-check: will this compile to KBC?
`ketoy_list`	Enumerate a category: composables, constructors, defaults, shapes, modifiers, tokens, capabilities, canvas, language.
`ketoy_capabilities`	Define, register, and wire capabilities. Room, DataStore, Retrofit, and navigation recipes.
`ketoy_viewmodel`	Typed ViewModels, local state, effects, lifecycle, and the limits.
`ketoy_setup`	Gradle, BOM, the `ketoy { }` DSL, Hilt, signing, and version caveats.
`ketoy_build`	Build the `.ktx` locally with Gradle, with advisories and gotchas.
`ketoy_debug`	Map a `KetoyBC:` error or crash to a fix.
`ketoy_cli`	The Cloud CLI for auth, apps, push, and rollback.
`ketoy_search`	Free-text search over everything.

How your agent uses it

The pattern is simple. Before writing UI, the agent checks support. Before building, it validates. When a build fails, it debugs.

Orient. Call ketoy_overview to learn the shape of what Ketoy allows.
Check. Before using a composable or API, call ketoy_check_support. The answer says whether it is catalogued, how it renders, and how to bridge it if not.
Validate. Paste the draft into ketoy_validate to catch forbidden APIs and uncatalogued composables before you compile.
Debug. When you hit a KetoyBC: error, ketoy_debug maps the symptom to the fix.

Next: ship what your agent builds with the Ketoy CLI.