Claude Code
Read a codebase, edit files and run commands from your terminal.
Recommended model
claude-opus-5-5Setup guide →ModelFold · AI Agents
Connect your AI coding agents with one ModelFold API key.
Connect Claude Code, Codex CLI, Grok Build, Trae and ModelFold skills. Find API base URLs, model IDs, setup guides and billing answers.
OpenAI · Chat Completions
https://modelfold.ai/v1Anthropic · Claude Code
https://modelfold.ai/anthropicChoose the protocol your tool expects. Claude Code uses the Anthropic base URL; Codex CLI uses the OpenAI Responses address in its setup guide.
Choose your coding tool
All integration guides →Setup documentation is available in English and Chinese.
Claude Code Desktop
Configure a desktop client to work with your ModelFold key.
Recommended model
claude-opus-5-5Setup guide →Claude Code IDE
Bring coding assistance into your editor with the documented IDE setup.
Recommended model
claude-opus-5-5Setup guide →Codex CLI
Plan changes, edit code and run checks with the Responses API.
Recommended model
gpt-6-1-solSetup guide →Grok Build
Connect the Grok coding agent through the xAI Responses protocol.
Recommended model
grok-4-7Setup guide →Trae
Use a custom OpenAI-compatible model for chat and agent workflows.
Recommended model
gpt-6-1-solSetup guide →modelfold-models skill
Help your agent discover ModelFold APIs and build apps with media models.
Recommended model
gpt-6-1-solSetup guide →modelfold-chat-agents skill
Give coding agents reusable instructions for connecting ModelFold chat models.
Recommended model
claude-opus-5-5Setup guide →What goes in the model ID field?
Use the exact API model ID, not a display name. If your tool cannot fetch the list, copy an ID from the model market and enter it manually. Protocol-specific lists may differ; follow the setup guide for your tool.
Browse models and IDs →Frequently asked questions
How do I connect Trae to a third-party API?
Add a custom OpenAI Chat Completions model with your ModelFold key, https://modelfold.ai/v1 and an exact model ID. If full-URL mode is enabled, use https://modelfold.ai/v1/chat/completions. Follow the Trae setup guide for the matching address mode.
Can Claude Code use other models?
Claude Code can connect to ModelFold through its Anthropic-compatible gateway. Set ANTHROPIC_BASE_URL to https://modelfold.ai/anthropic and choose a model supported by that gateway; an OpenAI-compatible model alone does not guarantee Claude Code compatibility.
Can I connect Cursor to ModelFold?
A dedicated Cursor integration has not been verified here. Before using https://modelfold.ai/v1, check whether your Cursor version and selected feature support a custom OpenAI-compatible base URL; chat compatibility does not establish Agent compatibility.
Which coding tools can I connect?
The guides cover Claude Code, desktop clients and IDE integrations, Codex CLI, Grok Build and Trae. ModelFold skills provide reusable setup instructions. Your tool must support a custom API address and a compatible protocol.
How does usage-based billing work?
Charges follow the selected model’s detail page and actual usage. Input, output and cache pricing can differ. Agents can make multiple requests for one task and include tool definitions and conversation history; check usage in your account.
What if my tool cannot load the model list?
Enter the model ID manually from the model market. GET /v1/models requires your ModelFold API key. For Anthropic, OpenAI Responses or xAI, use the model list specified in the tool’s guide.
How do I troubleshoot a 401 error?
Check that you are using an active ModelFold key and the documented authentication header. Verify that environment variables have not overridden the API address. Never share your full key in screenshots or logs.
How do I troubleshoot a 404 error?
Check the protocol, base URL and exact model ID. Avoid missing or repeated /v1 segments. Some tools append the endpoint automatically; others accept a complete URL. Follow the tool-specific guide.
Can every chat model use tools?
No. Choose a model with documented tool-calling support and a protocol compatible with your client. We do not recommend gemini-3-pro for agent workflows because its tool calling is unreliable. For Claude tools, we recommend claude-opus-5-5. In our Anthropic Messages test, tool_choice auto returned a tool call; the upstream rejected forced tool/any choices with HTTP 400.