Hologram

Hologram comes from building games. The bulk of the asset work runs through Blender into glTF: characters, props, weapons, and at some point an AI coding agent became a real part of that pipeline: it writes the Blender scripts, runs the exports, and rearranges the .glb files that ship. That was a huge speed-up, right up until the realisation that there was no real idea what it was doing. Assets changed, exports appeared, and reconstructing which step touched which file meant scrolling back through a terminal.

Hologram closes that gap. It tails a single event log and shows the agent's live activity: edits, shell commands, exports, right next to the assets those actions produce, in one local dashboard. Then it hands the agent that same pipeline back as a few MCP tools, so human and agent end up looking at the same picture instead of talking past each other.

Plenty of tools inspect a .glb. Hologram is the only one that puts a live feed of your agent's pipeline activity next to the assets it's producing: and gives the agent a render, so it can see an export, not just count its nodes.

Hologram is deliberately read-only / non-destructive: it observes, introspects, validates, and previews your pipeline, but it never modifies your assets. The one nuance is render_asset. It drives a live Blender to produce an image, in a throwaway scene with your scene restored afterward. Four tools are strictly read-only; the render is non-destructive. Checks you author run over assets but can't modify anything, and never run inside the MCP server.

The second constraint is architectural austerity: no framework, no build step, no database. A stdlib HTTP server, a JSONL event log, and pure-Python glTF parsing. The MCP server imports none of your project code, and Hologram itself never imports bpy. It drives Blender over a socket, so its import purity stays intact. Austerity here isn't minimalism for its own sake; it's what makes the tool trustworthy enough to point at the pipeline that produces the assets you actually ship.

Four pieces, none needing a manual install: an activity hook (a stdlib-only Claude Code hook that logs sessions, shell commands, edits, and MCP calls into the event log), the MCP server (launched by uvx per session), five guided skills (/hologram:start, inspect, check, status, create-skill, a natural-language front door), and the SSE dashboard you run in a terminal when you want eyes on the pipeline.

On top of the live feed: read-only checks you author in .hologram/checks.py, fingerprint-based regression diffing so the dashboard can answer "what changed since the last check", and pipeline_status. One MCP read of what's wrong right now.