Judgment

Judgment plugin

The official Judgment plugin for AI coding agents. One repo, installable in both Claude Code and OpenAI Codex, bundling:

1. The Judgment MCP server (https://mcp.judgmentlabs.ai): query traces, sessions, behaviors, judges, prompts, datasets, tests, automations, and agent memory directly from your editor.
2. The judgment-tracing skill: best-practice guidance for instrumenting apps with tracing, writing evaluations, building code judges, and running offline agent tests.
3. The mcp-server-best-practices skill: usage patterns for the Judgment MCP itself (search_traces batching, filter reference, fan-out queries).

The MCP gives your agent live data and actions; the skills teach it how to use Judgment well. Together they cover both "what's happening in my agent" and "how do I improve it."

Install

Claude Code

claude plugin marketplace add JudgmentLabs/judgment-plugin
claude plugin install judgment@judgment-plugin

If you previously added the Judgment MCP manually (for example claude mcp add judgment-mcp ...), remove it first so you do not end up with two Judgment servers:

claude mcp remove judgment-mcp

Codex

codex plugin marketplace add JudgmentLabs/judgment-plugin

Then open the plugin browser inside Codex with /plugins, select the judgment marketplace, and choose Install plugin. Start a new thread after installing, then authenticate:

codex mcp login judgment

Other agents (npx skills)

For Cursor, Windsurf, and other agents, install the skills with the skills CLI:

npx skills add JudgmentLabs/judgment-plugin

Then add the MCP server manually (see Standalone MCP below).

Authentication

The plugin ships with OAuth as the default (Judgment's recommended method). On first use your agent opens a browser window where you sign into Judgment and authorize access. No API keys or headers to manage. Organization selection happens per tool call via the organization_id argument (start with list_organizations).

API key alternative (headless / CI)

Claude Code — add an Authorization header to .mcp.json:

{
  "mcpServers": {
    "judgment": {
      "type": "http",
      "url": "https://mcp.judgmentlabs.ai",
      "headers": {
        "Authorization": "Bearer ${JUDGMENT_API_KEY}"
      }
    }
  }
}

Codex — configure the server in ~/.codex/config.toml instead of the plugin, using a bearer token from the environment:

[mcp_servers.judgment]
url = "https://mcp.judgmentlabs.ai"
bearer_token_env_var = "JUDGMENT_API_KEY"

Then export the key before starting your agent:

export JUDGMENT_API_KEY="your_key_here"

The Judgment MCP needs only the Authorization: Bearer header; there is no org-id header. Get your API key from the Judgment platform.

Standalone MCP (no plugin)

If you only want the MCP server without the skills, add it directly:

# Claude Code
claude mcp add judgment --transport http https://mcp.judgmentlabs.ai

# Codex (~/.codex/config.toml)
# [mcp_servers.judgment]
# url = "https://mcp.judgmentlabs.ai"

See the Judgment MCP docs for full details.

What's inside

judgment-plugin/
├── .claude-plugin/          # Claude Code plugin + marketplace manifests
├── .codex-plugin/           # Codex plugin manifest
├── .agents/plugins/         # Codex marketplace manifest
├── .mcp.json                # the Judgment MCP server (shared by both agents)
├── assets/                  # Judgment logo and composer icon
└── skills/
    ├── judgment-tracing/
    └── mcp-server-best-practices/

Claude Code namespaces the MCP tools as mcp__plugin_judgment_judgment__<tool> (for example mcp__plugin_judgment_judgment__search_traces); use the full name in permission rules or subagent tool lists. In Codex, invoke the skills explicitly with $judgment-tracing or $mcp-server-best-practices.

Local development

claude --plugin-dir .        # load this checkout for one session
claude plugin validate .     # schema-check

License

Apache-2.0