| name | install |
| description | Step-by-step playbook for installing mcp-blender-agent into an end user's agent harness (Claude Code / Copilot / Cursor / Claude Desktop). Read when the user asks to install the MCP or follow the install prompt. |
Install playbook (skill placeholder)
The full installer brain lives at install/AGENT-INSTALL.md. It is not finalized yet — it lands in Phase 5 of the roadmap.
When the user asks you to install blender-agent into their setup, follow install/AGENT-INSTALL.md end to end. Until that file exists in full form, use the temporary skeleton at install/INSTALL.md which lists manual steps.
Why this skill exists now (before Phase 5)
So that the installer surface area is reserved and the agent has somewhere to land when asked. The full content (detect → plan → ask → execute → inject → verify → uninstall) mirrors mcp-unreal-agent's install/AGENT-INSTALL.md.
High-level shape (preview)
- Detect — OS, Blender version, agent harness, conflicts.
- Plan — primary harness, addon install path, MCP config entry, managed instruction block.
- Ask — confirm with user.
- Execute — install addon, install TS server, merge MCP config.
- Inject — copy passive context skill into the harness, add managed block.
- Verify — call
server_status end-to-end.
- Uninstall — strip managed block, remove skill files, remove MCP entry, disable addon. Leaves
.blend files alone.
For Blender specifically (vs UE):
- No build step. The addon is a folder +
bl_info dict.
- Install via
blender --background --python-expr "bpy.ops.preferences.addon_install(filepath='...'); bpy.ops.preferences.addon_enable(module='BlenderAgent')".
- No source-control complications (no
.uproject to edit).
- Headless verification is trivial — same
--background invocation we use in CI.
