mcpField
C
ConvergeQA Compare (Due Diligence)

ConvergeQA Compare (Due Diligence)

Local
convergeqa.netAI & Agents0Updated 16w ago

About ConvergeQA Compare (Due Diligence)

ConvergeQA Compare is a multi-model document review system for due diligence that compares one prompt across models and provides public verification receipts. Its stdio MCP server lets clients submit jobs, check status, and retrieve review artifacts.

Key features

  1. 01Submit due-diligence comparisons
  2. 02Check job status
  3. 03Retrieve review packets
  4. 04Download review bundles
  5. 05Submit, poll, and fetch in one run

Use cases

  1. 01Compare multiple models’ responses to one prompt
  2. 02Review due-diligence results and receipts
  3. 03Download review artifacts for later examination

MCP support

It provides a stdio JSON-RPC MCP server exposing Compare due-diligence tools to submit jobs, check status, retrieve packets, and download bundles.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"convergeqa-compare": {
"command": "uvx",
"args": [
"convergeqa-mcp"
],
"env": {
"CONVERGEQA_API_KEY": "<CONVERGEQA_API_KEY>",
"CONVERGEQA_SERVICE_ACCOUNT_KEY": "<CONVERGEQA_SERVICE_ACCOUNT_KEY>"
}
}
}
}
Run commanduvx convergeqa-mcp
Environment variables to set
CONVERGEQA_API_KEYSecretConvergeQA Developer API key (from https://convergeqa.net/developers). Reviews consume paid credits.
CONVERGEQA_SERVICE_ACCOUNT_KEYSecretConvergeQA service-account key (alternative to the Developer API key).

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
pypiconvergeqa-mcp0.2.0

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project 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:

uvx convergeqa-mcp reviews
uvx convergeqa-mcp compare
pipx run convergeqa-mcp reviews
pipx run convergeqa-mcp compare

Or directly from Git:

uvx --from git+https://github.com/convergeqa/convergeqa-mcp.git convergeqa-mcp-reviews
uvx --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.

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

Add Compare as a second server:

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

Claude Desktop

Example claude_desktop_config.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):

[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.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.