GripForge Studio vs CLI vs MCP: Which Tool Fits Your Asset Task

Use Studio when you need to look at an asset and judge it — socket alignment, tiling seams, silhouette, terrain texture scale. Use the CLI (grip-forge) when you want to run your chosen local agent with a connected GripForge workspace and follow its generation jobs. Use the hosted MCP endpoint (https://gripforge.ai/mcp) when a coding agent or hosted chat client needs to call GripForge tools remotely — it returns asset URLs, not local files. Use the local MCP client (@gripforgeai/mcp) when you want an agent to pull library assets directly into your working folder without going through the CLI.

All three are complementary, not competing options for the same job. One task can use visual inspection, local agent work and remote tool calls at different stages.

Decision matrix

Practical steps

1. Start by classifying the task, not the tool. Ask: do I need to see something, do I need a repeatable job I can rerun, or do I need an agent to fetch a file into this exact folder? That answer picks the tool for you.

2. Visual inspection → Studio. Open the relevant studio (equipment, character/animation, texture, terrain, or map) from gripforge.ai, load the candidate asset, and compare it against your reference — for example, checking socket alignment before a transfer to an engine, per the Unreal socket transfer checklist. Review the actual appearance before approving it for the next stage. A completed generation job alone does not establish that the result meets the visual requirements.

3. Local agent work → CLI. Follow the official CLI setup instructions. The CLI needs Node.js 22.19 or later and a separately installed supported agent. Browser authorization connects your terminal to a chosen GripForge workspace; the agent retains its normal local file permissions.

Use the CLI to follow generation jobs, return to the selected agent's conversation, or switch supported agents while preserving the GripForge login and workspace. A job reaching completion establishes that its server-side work finished. Confirm separately that any required asset has been downloaded and integrated into the local project.

The agent or model provider uses your own account, while GripForge generation uses workspace credits. GripForge AI is listed as coming soon, not a currently available hosted-agent subscription.

4. Hosted tool integration → hosted MCP. If your coding agent or hosted chat client supports MCP servers, point it at https://gripforge.ai/mcp with an x-api-key header (see MCP docs for the current tool list and auth setup). Call the tool for the asset type you need. Inspect the result appropriate to the tool: it may be structured data, a job reference or an asset URL. For asset downloads, the hosted endpoint supplies a URL rather than writing to your local filesystem; your agent or another local download step retrieves the file.

5. Pulling assets into your repo → local MCP client. Install @gripforgeai/mcp and let your agent call gripforge_library_pull. This client runs locally, so it can write the downloaded asset into your actual project directory instead of just handing back a link. This is the step that closes the gap left by the hosted endpoint — use it when you want "agent asks, file appears in my repo" behavior.

Troubleshooting

• CLI won't start / version errors — check node -v against the 22.19+ requirement before anything else.
• Agent commands not recognized — the agent (e.g., Claude Code) must be installed separately from grip-forge; follow the connection command for that supported agent in the CLI documentation after installing it.
• Job never finishes in jobs watch — re-check with grip-forge jobs for its current state, and confirm the selected workspace using the CLI account/workspace controls.
• Hosted MCP tool returns a URL but nothing lands in your project — that's expected; the hosted endpoint doesn't have local file access. Either fetch the URL manually or switch that step to the local MCP client's gripforge_library_pull.
• Switched agents and lost context — use grip-forge switch to change agents while keeping login/workspace state, and grip-forge resume to continue that agent's last conversation instead of starting over.
• Unsure whether MCP or CLI is "newer" — they're not sequential versions of each other; MCP remains available alongside the CLI, and one doesn't disable the other.

Quick checklist before you pick a tool

• Am I judging how something looks? → Studio.
• Do I need this job to be rerunnable and trackable by ID? → CLI.
• Is a hosted agent calling GripForge remotely and fine with fetching a URL itself? → Hosted MCP.
• Do I need the file to land in my repo without a manual fetch step? → Local MCP client.

For the level-design stage of a project, see From Level Plan to Playable Greybox with GripForge. For the full command reference and setup, see the CLI page and general docs.

FAQ

Does the CLI replace MCP?

No. They run alongside each other — the CLI is for local, repeatable, agent-of-your-choice jobs; MCP is for tool calls from coding agents or hosted clients.

Can the hosted MCP endpoint write files into my project?

No. Tool responses can contain data, job references and asset URLs, but the hosted server does not write to your local project. A local agent or client must retrieve and write the files.

What do I need installed before running grip-forge commands?

Node.js 22.19 or later, plus your preferred agent installed separately. Full steps are on the /cli page.

Is GripForge AI available as an agent option yet?

Not yet — it's listed as coming soon, so connect an agent you already use in the meantime.