arize-phoenix: v20.20.0
Arize Phoenix v20.20.0 expands MCP tooling with GraphQL capabilities, allowlists core tables for MCP SQL queries, introduces a Docker Sandboxes provider, and trims dependencies by removing SciPy and scikit-learn.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
Arize Phoenix version 20.20.0 introduces several enhancements for agent and MCP integration. New features include GraphQL schema, query, and mutation tools for MCP, as well as an allowlist for SQL tables (annotation, evaluator, prompt, split, and job). A Docker Sandboxes provider was added alongside a prompt hill-climb task on text-to-SQL for Harbor benchmarks. Notably, runtime dependencies were trimmed by removing SciPy and scikit-learn, and pydantic-ai was upgraded to 2.52 with the Anthropic max tokens override removed. PXI routing was optimized to direct data questions to analytics SQL, reducing wasted turns.
可执行摘要
Arize Phoenix v20.20.0 adds MCP GraphQL schema/query/mutation tools and table allowlisting for SQL querying. It also drops SciPy and scikit-learn runtime dependencies, adds a Docker Sandboxes provider, updates pydantic-ai to 2.52, and refines PXI routing to analytics SQL.
- Agent 实用度
- 77/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Verify environments do not rely on SciPy or scikit-learn being transitively provided by arize-phoenix runtime dependencies.
- Update MCP client configurations if intending to leverage newly added GraphQL query/mutation tools or allowlisted SQL tables.
- Review Anthropic model usage in the playground due to the addition of claude-haiku-5-5 and removal of retired models.