ConvergeQA Compare (Due Diligence)

io.github.convergeqa由 GitHub 账号 convergeqa 发布v0.2.0更新于 Oct 24, 2026

Compare one prompt across a multi-model panel side by side, with public verification receipts.

已验证STDIO仅桌面数据与分析AI 与机器学习

概览

AI 生成的概览

一个本地 MCP 服务器,把提示提交给 ConvergeQA 的多模型尽职调查评审面板,并取回生成的包、捆绑包与验证凭证。

可以做什么
提交一个私有的 Compare 尽职调查任务并轮询其状态直至完成。以 JSON 形式获取结果包,可选择写入指定路径。将完整的 Compare 捆绑包下载到本地文件。通过一次调用完成提交、轮询、获取与下载的整个流程。
适用场景
当你希望由多个模型并排评审同一提示或文档,并把分歧与证据保存为可检索的产物,而不是只得到一个模型的回答时适用。
运行要求
通过 uvx 或 pipx 从 PyPI 包 convergeqa-mcp 以 stdio 在本地运行,需要 Python 3.10 或更高版本。需要通过 CONVERGEQA_API_KEY 或 CONVERGEQA_SERVICE_ACCOUNT_KEY 环境变量提供 ConvergeQA 开发者 API 密钥或服务账号密钥,并需要访问 ConvergeQA 的网络连接。
安装前请注意
评审会消耗付费额度,每个任务都可能产生费用。服务器环境中需要 ConvergeQA API 或服务账号凭据;README 称该工具不存储密钥。任务会提交到 ConvergeQA 的远程服务,因此提示内容会离开本机。部分工具会把结果包或捆绑包文件写入调用方指定的路径。

安装

在 SourceWeft 中

  1. 打开 控制台中的 ConvergeQA Compare (Due Diligence),将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

convergeqa-mcp

ConvergeQA is a multi-model document review system with public verification receipts. Its panels challenge thinking, surface disagreements, and document review evidence, while the responsible individual decides what to accept, reject, or revise.

This package exposes two stdio MCP servers:

  • convergeqa-mcp reviews (or convergeqa-mcp-reviews): Critique and Iterate review tools.
  • convergeqa-mcp compare (or convergeqa-mcp-compare): Compare due-diligence review tools.

Both servers delegate to the same CLI client contracts used by ConvergeQA agent tooling. Reviews consume paid credits. The servers read a ConvergeQA Developer API key or service-account key from environment variables at runtime, and the tool never stores keys.

Links

Install

From PyPI:

bash
uvx convergeqa-mcp reviewsuvx convergeqa-mcp compare
bash
pipx run convergeqa-mcp reviewspipx run convergeqa-mcp compare

Or directly from Git:

bash
uvx --from git+https://github.com/convergeqa/convergeqa-mcp.git convergeqa-mcp-reviewsuvx --from git+https://github.com/convergeqa/convergeqa-mcp.git convergeqa-mcp-compare

Claude Code

Set CONVERGEQA_API_KEY or CONVERGEQA_SERVICE_ACCOUNT_KEY before starting your MCP client. The default key names are uppercase environment variables.

bash
claude mcp add convergeqa-reviews --env CONVERGEQA_API_KEY=$CONVERGEQA_API_KEY -- uvx convergeqa-mcp reviews

Add Compare as a second server:

bash
claude mcp add convergeqa-compare --env CONVERGEQA_API_KEY=$CONVERGEQA_API_KEY -- uvx convergeqa-mcp compare

Claude Desktop

Example claude_desktop_config.json:

json
{  "mcpServers": {    "convergeqa-reviews": {      "command": "uvx",      "args": ["convergeqa-mcp", "reviews"],      "env": {        "CONVERGEQA_API_KEY": "${CONVERGEQA_API_KEY}"      }    },    "convergeqa-compare": {      "command": "uvx",      "args": ["convergeqa-mcp", "compare"],      "env": {        "CONVERGEQA_API_KEY": "${CONVERGEQA_API_KEY}"      }    }  }}

If your MCP client does not expand environment placeholders, set the key in the client environment by the method that client supports.

Other MCP Clients

These are standard stdio JSON-RPC MCP servers, so any MCP-capable client can run them: Codex CLI, OpenCode, Cursor, OpenClaw, Hermes Agent, and others. The pattern is always the same: launch the command, pass the API key in the environment.

Codex CLI (~/.codex/config.toml):

toml
[mcp_servers.convergeqa-reviews]command = "uvx"args = ["convergeqa-mcp", "reviews"]env = { "CONVERGEQA_API_KEY" = "<your key>" }
[mcp_servers.convergeqa-compare]command = "uvx"args = ["convergeqa-mcp", "compare"]env = { "CONVERGEQA_API_KEY" = "<your key>" }

For clients without a uvx runtime, install once with pipx install convergeqa-mcp and point the client at the installed convergeqa-mcp-reviews and convergeqa-mcp-compare commands. Python 3.10 or newer; no other runtime dependencies.

Tools

convergeqa-mcp reviews (16 tools):

  • convergeqa_critique_start: Start a private authenticated Critique agent review.
  • convergeqa_critique_status: Fetch owner-scoped Critique job status.
  • convergeqa_critique_packet: Fetch owner-authenticated Critique packet JSON and optionally write it to a path.
  • convergeqa_critique_session: Fetch owner-authenticated Critique session packet.
  • convergeqa_critique_sessions: List owner-authenticated Critique sessions.
  • convergeqa_critique_templates: List API-visible ConvergeQA templates for the configured credential.
  • convergeqa_critique_decide: Submit delegated Critique decisions.
  • convergeqa_critique_export: Download owner-authenticated Critique export ZIP to a caller-selected path.
  • convergeqa_iterate_start: Start a private authenticated Iterate agent review.
  • convergeqa_iterate_status: Fetch owner-scoped Iterate job status.
  • convergeqa_iterate_packet: Fetch owner-authenticated Iterate packet JSON and optionally write it to a path.
  • convergeqa_iterate_session: Fetch owner-authenticated Iterate session packet.
  • convergeqa_iterate_sessions: List owner-authenticated Iterate sessions.
  • convergeqa_iterate_templates: List API-visible ConvergeQA templates for the configured credential.
  • convergeqa_iterate_decide: Submit delegated Iterate decisions.
  • convergeqa_iterate_export: Download owner-authenticated Iterate export ZIP to a caller-selected path.

convergeqa-mcp compare (5 tools):

  • convergeqa_compare_submit: Submit a private authenticated Compare due-diligence job.
  • convergeqa_compare_status: Fetch owner-scoped Compare due-diligence job status.
  • convergeqa_compare_packet: Fetch owner-authenticated packet JSON and optionally write it to a caller-selected path.
  • convergeqa_compare_bundle: Download owner-authenticated Compare due-diligence bundle to a caller-selected path.
  • convergeqa_compare_run: Submit, poll, fetch packet, and optionally download packet/bundle artifacts.

Credentials

The review servers accept these default environment variable names:

  • CONVERGEQA_API_KEY
  • CONVERGEQA_SERVICE_ACCOUNT_KEY

Tool arguments may override the environment variable name, but literal credential values are rejected.

Protocol

The current servers speak stdio JSON-RPC and report MCP protocol version 2024-11-05, matching the existing ConvergeQA MCP server behavior.

来源:README.md,提交 a80c1f8

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.2.0最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分