Connect AI tools to the Kimchi API through the Kimchi MCP server.
The Kimchi Model Context Protocol (MCP) server connects AI tools to the Kimchi API. Any MCP client can connect. Supported tools include Cursor, VS Code, Windsurf, JetBrains, Claude Desktop, ChatGPT, Kimchi Studio, Claude Code, Gemini CLI, and Codex.
What is MCP?
Model Context Protocol (MCP) is an open standard that allows AI applications to securely access external data sources and tools. Once connected, your AI tool can call:
| Tool | What it does |
|---|---|
list-endpoints | Returns all API paths with their HTTP methods and summaries. |
get-endpoint | Returns the full definition of one endpoint: parameters, security, and servers. |
search-endpoints | Searches paths, operations, and schemas by keyword. |
get-server-variables | Returns the server variables defined in the Kimchi OpenAPI spec. |
execute-request | Makes a live API call to the Kimchi API and returns the response. |
Kimchi hosts a remote MCP server at:
https://docs.kimchi.dev/mcpSet up with an agent
Paste this prompt into your AI tool to let it configure the connection:
Set me up with the Kimchi MCP server in this AI tool.
First understand what client you are running in, for example Cursor, Windsurf, Claude Desktop, or Gemini CLI. Then read the setup instructions here: https://docs.kimchi.dev/reference/mcp and follow the section for that client to write the configuration. If the client is not listed, use the server URL and the X-API-Key header documented on that page to write the configuration in the client's native format.
Ask me for my Kimchi API key and wait for my answer. If I do not have a key, send me to https://app.kimchi.dev/settings to create one, then continue.
After setup, list the MCP server's tools to confirm the connection. If the client needs a restart or reload, tell me exactly what to restart and resume this conversation after I restart it. Do not start a new conversation.Set up manually
Configure your AI development tools to connect to the Kimchi MCP server at:
https://docs.kimchi.dev/mcpThe examples use the X-API-Key header. To use a bearer token instead, see Kimchi API key.
API keys in config filesClient config files store your API key in plain text. Keep them out of shared repositories and source control. Rotate your API key if a config file leaks.
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"kimchi": {
"url": "https://docs.kimchi.dev/mcp",
"headers": {
"X-API-Key": "<your-kimchi-api-key>"
}
}
}
}Selecting an API server
The Kimchi MCP server URL accepts query parameters that pin the API server and override server variables defined in the Kimchi OpenAPI spec. This gives your AI tool a deterministic target. Your tool connects to exactly the server you specify, not to a server the AI selects at runtime.
URL parameters
| Parameter | Example | Description |
|---|---|---|
server | ?server=0 | Select a server from the Kimchi OpenAPI spec by zero-based index. The MCP server exposes only that server and rejects requests that target other servers. |
server_variables | ?server_variables=region=eu-central-1 | Override a server variable defined in the spec. Format: KEY=VALUE. Repeat the parameter for multiple variables. |
Example
The Kimchi API spec defines three servers:
| Index | Server | Description |
|---|---|---|
0 | https://api.cast.ai/ | US endpoint |
1 | https://api.eu.cast.ai/ | EU endpoint |
2 | https://api.india.cast.ai/ | India endpoint |
To connect the MCP server to a specific endpoint, select it below:
{
"mcpServers": {
"kimchi": {
"url": "https://docs.kimchi.dev/mcp?server=0",
"headers": {
"X-API-Key": "<your-kimchi-api-key>"
}
}
}
}Append server and server_variables directly to the MCP URL.
Authentication
Connecting to the Kimchi MCP server needs no authentication. The tools need your Kimchi API key when they call the Kimchi API on your behalf. Configure one API key header in your client:
| Header | Value | Purpose |
|---|---|---|
X-API-Key or Authorization | Your Kimchi API key in one of two formats | Authenticates against the Kimchi API |
The MCP server forwards this header to the Kimchi API when a tool calls the API on your behalf.
Kimchi API key
You can pass your Kimchi API key in two formats:
| Format | Header |
|---|---|
| API key | X-API-Key: <your-kimchi-api-key> |
| Bearer token | Authorization: Bearer <your-kimchi-api-key> |
Both formats use the same Kimchi API key and provide identical access. Use whichever format your tooling or organization requires.
Claude Desktop limitationBearer token auth is not supported on Claude Desktop via the
mcp-remotebridge. Use theX-API-Keyheader for Claude Desktop configurations.
Obtaining an API key
Generate an API key in the Kimchi console at Kimchi → API Keys. For details, see API Keys and Service Keys.
Testing your MCP setup
Once configured, test your MCP server connection:
-
Open your AI editor, for example Cursor, Windsurf, Gemini CLI, or Kimchi Studio.
-
Start a new chat with the AI assistant.
-
Ask about Kimchi, for example:
List the Kimchi API endpoints you can call.
The AI should now have access to the Kimchi API and documentation through the MCP server.
Troubleshooting
If the MCP server is not responding or returning errors, check the MCP server logs for your client before investigating further.
Open the Output panel in Cursor (View > Output) and select MCP from the dropdown. Logs show connection status and any errors returned by the server.
After reviewing logs, also verify that:
- Your API key is valid and has not been deleted or disabled
- The header value in your config does not include extra spaces or line breaks
X-API-Key(orAuthorization) is present in your configuration- For Claude Desktop,
--headervalues have no space after the colon: useX-API-Key:<key>, notX-API-Key: <key> - You fully quit and relaunch your AI editor after updating the config file. Most clients do not reload the MCP configuration without a restart