菜单
Use the official Braintrust MCP server to search Braintrust docs, inspect projects, query logs and experiments, summarize eval results, and generate permalinks.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/braintrustdata/braintrust-codex-plugin --skill braintrust命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | braintrust |
| description | Use the official Braintrust MCP server to search Braintrust docs, inspect projects, query logs and experiments, summarize eval results, and generate permalinks. |
Use this skill when the user asks to work with Braintrust data, documentation, experiments, datasets, prompts, logs, traces, evals, or SDK setup.
Use the braintrust MCP server.
Default US data plane URL:
https://api.braintrust.dev/mcp
EU data plane URL:
https://api-eu.braintrust.dev/mcp
Self-hosted URL:
Use the organization's configured Braintrust API URL with /mcp.
Prefer the plugin-provided OAuth flow for the remote MCP server.
Braintrust also supports API-key bearer authentication. In Codex TOML, the verified configuration is:
[mcp_servers.braintrust]
url = "https://api.braintrust.dev/mcp"
bearer_token_env_var = "BRAINTRUST_API_KEY"
Never print or expose the API key.
Use search_docs for Braintrust documentation questions.
Use docs://sdk-install before search_docs when the user asks to install Braintrust, set up tracing, add observability, or configure an eval.
Use resolve_object to convert Braintrust names, IDs, and URLs into object metadata.
Use list_recent_objects to discover accessible projects, experiments, datasets, prompts, functions, and other recent objects.
Use infer_schema before writing SQL when the available fields are unclear.
Use sql_query to query experiments, datasets, project logs, traces, and summaries.
Use summarize_experiment for aggregate experiment metrics and baseline comparisons.
Use generate_permalink when the user needs a shareable Braintrust link.
For production debugging, first resolve the relevant project or object, infer schema if needed, query logs with sql_query, then generate permalinks for important traces.
For experiment analysis, list or resolve the experiment, summarize it, compare against a baseline when available, then use SQL for row-level examples.
For documentation and setup, read docs://sdk-install for SDK installation tasks and use search_docs for follow-up details.
Use MCP when the user wants conversational exploration of Braintrust data or docs.
Prefer the bt CLI for repeatable workflows, CI gates, running evals, scripted operations, and setup flows that should be deterministic.
The Braintrust MCP is currently best treated as read-oriented. Do not invent write capabilities.