MeshCrunch MCP

3D optimization inside your AI workflow.

Point an AI client at one URL and it can work with the assets already in your account — simplify, generate LOD chains, convert formats — on your existing sign-in and plan.

https://mcp.meshcrunch.com/mcp

Works with any client that supports a remote MCP server over Streamable HTTP with OAuth. Tested with Claude Code. Available on Starter and Pro plans.

Set up your AI client

Claude Code

claude mcp add --transport http meshcrunch https://mcp.meshcrunch.com/mcp
/mcp

Codex CLI

codex mcp add meshcrunch -- https://mcp.meshcrunch.com/mcp
codex mcp login meshcrunch

Gemini CLI

gemini mcp add --transport http meshcrunch https://mcp.meshcrunch.com/mcp

Either way a browser opens for you to sign in and approve the two permissions below. There is no API key to create, copy, or leak.

What the agent can do

“What's in my library?” · “What would four LOD levels cost on the turbine housing?” · “Simplify it to 20% and tell me when it's done.” A tool that creates a job spends processing credits at exactly the rate the same operation costs in Studio, scaled by the size of the file — which is why an agent can price a call before making it.

ToolWhat it doesCost
list_assetsList the assets in your library.Free
get_assetRead one asset's format, size and geometry counts.Free
get_asset_download_urlMint a short-lived download URL for a source asset.Free
list_jobsList recent runs and their status.Free
get_jobPoll one run until it is done or failed.Free
get_job_resultRead a finished run's measurements and download its output.Free
get_account_usageRead the plan's allowances, what is used and when it resets.Free
estimate_operationPrice a call before making it. Writes nothing, reserves nothing.Free
simplify_assetReduce an asset's triangle count.Spends credits
generate_lodsGenerate a level-of-detail chain.Spends credits
convert_assetWrite an asset out in another format.Spends credits

Uploading is not a tool. Assets are added through the web app and mesh data never passes through the MCP server. There is also one resource, meshcrunch://config, holding the limits and formats the service enforces.

Looking for workflow ideas? Browse the MCP prompt library — copyable prompts for inspecting, pricing and optimizing assets.

What you approve

  • mcp:read

    See your assets, history and usage, and price an operation before it runs.

  • mcp:writeCan spend credits

    Run simplify, LOD and conversion jobs, which spend your processing credits.

Every call is checked against the same ownership rules the web app passes — an agent reaches your assets and nobody else's. More on the security page.

Questions

Does this only work with Claude?
No. This is a standard remote MCP server rather than a per-vendor integration, so it works with any client that supports a remote MCP server over Streamable HTTP with OAuth. Every client reaches the same tools through the same sign-in; only the setup line differs. Claude is the one we have connected and verified ourselves, which is why it is marked as tested — a client we have not run is labelled, not described as supported.
Which plans include it, and does an agent get its own quota?
Starter and Pro. A free account can sign in successfully and is then refused with paid_plan_required, because entitlement is checked on every call rather than only when you connect — which also means access ends when a subscription lapses. An agent has no separate allowance: it resolves to the same account, on the same plan, through the same gates a browser passes, so connecting one raises no limit.
Can I connect an agent that cannot spend anything?
Yes, and it is a reasonable thing to do. Approve only the read permission: every read tool keeps working, pricing an operation still works, and anything that would cost you something is refused with forbidden.
Is there a REST API or an API key instead?
Programmatic access is in beta on paid plans through the MeshCrunch MCP server, which agent clients connect to with your MeshCrunch login; assets are still uploaded through the web app. There is no published REST reference, no SDK, no webhooks, and no self-service API keys.