Regression testing for AI agents: snapshot behavior, diff tool calls and catch regressions in CI.
Setup
Choose your client, copy the config, then fill in any required environment variables.
claude mcp add evalview --env OPENAI_API_KEY='<OPENAI_API_KEY>' -- uvx evalview~/.codex/config.toml
[mcp_servers.evalview]
command = "uvx"
args = ["evalview"]
env = { OPENAI_API_KEY = "<OPENAI_API_KEY>" }
# codex mcp add evalview --env OPENAI_API_KEY='<OPENAI_API_KEY>' -- uvx evalview~/.cursor/mcp.json
{
"mcpServers": {
"evalview": {
"command": "uvx",
"args": [
"evalview"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>"
}
}
}
}.vscode/mcp.json
{
"servers": {
"evalview": {
"type": "stdio",
"command": "uvx",
"args": [
"evalview"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>"
}
}
}
}~/.gemini/settings.json
{
"mcpServers": {
"evalview": {
"command": "uvx",
"args": [
"evalview"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>"
}
}
}
}claude_desktop_config.json
{
"mcpServers": {
"evalview": {
"command": "uvx",
"args": [
"evalview"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>"
}
}
}
}.trae/mcp.json
{
"mcpServers": {
"evalview": {
"command": "uvx",
"args": [
"evalview"
],
"env": {
"OPENAI_API_KEY": "<OPENAI_API_KEY>"
}
}
}
}Environment variables
OPENAI_API_KEY
Replace the placeholders with your own credentials and keep them out of version control.