Crimp for AI agents · public beta
Let your AI look up real components, calculate a wire or DC-DC charger circuit, and build a small electrical plan. Crimp does the calculations. You inspect the assumptions and edit the drawing.
Free public tools. No Crimp account or API key required. Your AI client’s own access rules apply.
1 · Connect
Direct setup is available now. Crimp is not yet listed in the public plugin directories; connect it below while directory publication is pending.
One public MCP server · Streamable HTTP · no authentication
https://crimp.studio/api/mcpCreate a personal plugin if custom MCP connections are enabled for your account or workspace.
@. Try a request below.Add this entry to ~/.codex/config.toml, keeping your other settings:
[mcp_servers.crimp]
url = "https://crimp.studio/api/mcp"Start a new Codex session. In the CLI, use /mcp to check the connection. The IDE extension also has an MCP server settings screen: select Streamable HTTP and enter the same URL.
Run this in the project where you want Crimp available:
claude mcp add --transport http crimp https://crimp.studio/api/mcpOpen Claude Code and use /mcp to check that Crimp is connected. No Crimp login is needed.
In a client that supports remote MCP, add the server URL above using Streamable HTTP and no authentication.
For an agent without MCP support, the same tools are available through the HTTP API.
The optional Crimp skill explains which tool to use and how to preserve assumptions. Reading the skill alone does not connect the tools.
Download the connection metadata and Crimp skill together. Each archive includes a README for manual setup. These are downloadable packages, not public directory installations.
The portable package uses the OpenAI Agent Plugins format. The Claude Code package uses Claude’s plugin format. Both call the same hosted Crimp tools.
2 · Ask
After connecting, ask your AI to use Crimp. It should ask for missing installation facts and report assumptions alongside the result.
Calculate a circuit
“Use Crimp to size the wire for my 12 V fridge circuit. Ask for the load and installation details first, then compare 3 m and 5 m one-way runs.”
Build a draft
“Use Crimp to plan a DC-DC charger circuit for my boat. Ask for the batteries, charger, charging limits and cable lengths. Give me a draft link with its assumptions.”
Check a system
“Use Crimp to analyze this exported Crimp project. Explain the highest-priority findings and what installation evidence is still missing.”
Provide the exported project file to your agent. Crimp’s tools cannot read your account’s saved projects.
3 · Review
A draft result includes a review link. Open it to see the components, assumptions and current checks, then choose whether to create an editable local project in Crimp.
Drafts support up to 10 counted equipment components. The tools apply Crimp’s supported ABYC E-11 calculations and checks; a result is planning assistance, not installation approval or compliance certification.
For developers
The calling agent supplies facts and topology; Crimp computes the results. Calls do not access saved projects or run a language model. API version 1.
GET /api/agent-tools returns the current tool descriptions and JSON input schemas. Use discovery as the authority for accepted fields.
curl https://crimp.studio/api/agent-tools
curl https://crimp.studio/api/agent-tools \
-H 'Content-Type: application/json' \
-d '{"tool":"search_components","arguments":{"query":"Orion"}}'Each POST accepts one {tool, arguments} object. Successful calls return {ok:true, tool, metadata, data}; tool errors return {ok:false, tool, metadata, error}. Metadata identifies the API, engine and project schema versions. HTTP-level rejection may return a transport error instead.
| Tool | Use it for |
|---|---|
search_components | Find catalog IDs for a part or category. |
get_component | Retrieve a component’s specifications, terminals, declared configuration and catalog provenance. |
calculate_circuit | Size a wire or DC-DC charger using the calculator’s explicit installation inputs. |
analyze_system | Review a supplied Crimp system graph and inspect circuit findings. |
create_draft | Build a catalog-based topology, analyze it and return a link to review it in Crimp. |
create_draft accepts {draft:{name, components, connections, assumptions}}. Component entries use real catalog IDs and explicit estimated flags. Connections identify component refs, ports, polarity and one-way cable length in metres. The discovery schema describes every required field.
The returned reviewUrl carries a bounded draft in its URL fragment. Opening it shows the components, assumptions and current checks before you choose to create a new local project. Your existing project is saved before switching. No cloud project or public share is created by the API.
The link contains the draft: treat it as information you are sharing with its recipient. Crimp removes the fragment before starting browser analytics and keeps a bounded copy in this tab’s session storage so a reload can recover the review. Importing or dismissing clears that copy. The draft is recalculated when opened; an incompatible engine-version link must be recreated.
The public endpoints emit bounded operational records containing the tool name, MCP or HTTP transport, success or error category, and duration. These records do not contain tool arguments, results or caller identifiers. Draft review links include a coarse transport label, and the app’s existing analytics records review opens and imports without the draft content. The label is not a user or request identifier.
Requests and graphs are bounded; HTTP request bodies are limited to 256 KiB and draft handoff payloads to 64 KiB. Best-effort rate limits protect the public endpoints; these are not a strict global quota or a service-level guarantee. Respect 429 and Retry-After, and retry transient failures with backoff. Correct invalid input before retrying.
Send only the project data needed for the requested calculation. Existing project and sharing credentials are neither required nor supported by this API.
Email [email protected] with your AI client, the setup step and any error message. For tool failures, include the tool name and a small example without private project data.