Editor Tools Manifest

Tissue Studio is gradually moving the in-page AI from prompt-only structured commands toward MCP-style tools.

Current status:

transitional MCP-compatible manifest

The first read-only inspect tools are now available:

getProjectSummary
getSceneObjects
findObjects
getScript
getPrefab
getAssets
getProjectLint

These tools let the AI inspect exact project state before mutating it.

Prefab child objects have a dedicated mutation tool:

updatePrefabObject

Use it instead of updateObject when modifying tags or components inside a prefab asset.

Inspect results are retained inside the current AI task as recentInspections, so a later step can use exact script code, object ids, prefab child names, tags, and lint results instead of guessing.

The existing backend command runner is still used for execution, but tool names, descriptions, rules, and input schemas are now described in a machine-readable manifest.

Why this exists

• Give AI a clearer tool contract.
• Avoid duplicating command lists in prompt, docs, and validators.
• Prepare for exposing the same project-scoped operations as real MCP tools later.
• Keep permissions scoped to the current project.

Runtime endpoint

GET /api/ai/tools/manifest

The endpoint returns the current editor tool manifest.

Safety rules

• Tools are scoped to the current project only.
• Tools must not access server files, admin data, other users, raw SQL, deployment, or system APIs.
• Execute one mutating tool per AI step.
• Use read-only inspect tools before risky mutations.
• Inspect and validate before claiming completion.
• Use reportToolIssue when an editor/runtime capability is missing.

Relationship to MCP

This is not a full MCP server yet. It is the first migration layer:

existing command runner -> manifest-described tools -> future MCP server/tools

The goal is to make tool semantics stable before changing transport/protocol.