Your prompts need tests too. Run prompts against real datasets, score outputs with LLM judges, version everything, and compare runs to see what got better.
Setup
Choose your client, copy the config, then fill in any required environment variables.
claude mcp add --transport http completionkit https://completionkit.com/mcp --header 'Authorization: <AUTHORIZATION>'~/.codex/config.toml
[mcp_servers.completionkit]
url = "https://completionkit.com/mcp"
http_headers = { Authorization = "<AUTHORIZATION>" }
# codex mcp add completionkit --url https://completionkit.com/mcp~/.cursor/mcp.json
{
"mcpServers": {
"completionkit": {
"url": "https://completionkit.com/mcp",
"headers": {
"Authorization": "<AUTHORIZATION>"
}
}
}
}.vscode/mcp.json
{
"servers": {
"completionkit": {
"type": "http",
"url": "https://completionkit.com/mcp",
"headers": {
"Authorization": "<AUTHORIZATION>"
}
}
}
}~/.gemini/settings.json
{
"mcpServers": {
"completionkit": {
"httpUrl": "https://completionkit.com/mcp",
"headers": {
"Authorization": "<AUTHORIZATION>"
}
}
}
}claude_desktop_config.json · Customize › Connectors › Add custom connector
{
"mcpServers": {
"completionkit": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://completionkit.com/mcp",
"--header",
"Authorization:<AUTHORIZATION>"
]
}
}
}.trae/mcp.json
{
"mcpServers": {
"completionkit": {
"url": "https://completionkit.com/mcp",
"headers": {
"Authorization": "<AUTHORIZATION>"
}
}
}
}