mergescope 0.1.0

Audit merged PRs against Jira tickets using a LangGraph agent and MCP tools

MergeScope

Audits merged PRs against Jira tickets. Extracts Jira IDs from PR titles, descriptions, branch names, and commit messages, then validates each ticket's fixVersion against your expected release.

Powered by a LangGraph agent that talks to GitHub and Jira through MCP servers. Works standalone or with Amazon Q.

Install

uvx mergescope --help

Or permanently:

uv tool install mergescope

Usage

mergescope \
  -f 2026-03-01 \
  -t 2026-03-31 \
  -v "1.2.0" \
  -r company/backend

Flag	Description
-f, --from-date	Start date (YYYY-MM-DD)
-t, --to-date	End date (YYYY-MM-DD)
-v, --fix-version	Expected Jira fixVersion
-r, --repo	GitHub repo (owner/repo)
-c, --config	Config file (default: mergescope.yaml)
--mcp-config	Amazon Q MCP config path
--llm-provider	anthropic, bedrock, copilot, or local
--llm-model	Model ID
--llm-base-url	API base URL (copilot/local)
--language	Response language (e.g. Spanish)
--verbose	Debug logging

Multilingual prompts

Set a custom prompt in any language via mergescope.yaml:

language: "es"
response_language: "Spanish"
prompt_template: |
  Necesito auditar tickets mergeados entre {from_date} y {to_date}.
  La fixVersion esperada es {fix_version}.
  Extrae los IDs de JIRA desde los merge commits del repositorio {repo}
  y valida su fixVersion.

Configuration

Add a mergescope.yaml in your project root:

repo: "nice/java-project"
jira_project_prefix: "PROJ"
jira_base_url: "https://company.jira.com"

llm:
  provider: "anthropic"       # anthropic, bedrock, copilot, local
  model: "claude-sonnet-4-6-20250620"

mcp:
  servers:
    github:
      name: "github-mcp-server"
    jira:
      name: "atlassian-mcp-server"

LLM providers

Provider	Model example	Notes
anthropic	claude-sonnet-4-6-20250620	Uses ANTHROPIC_API_KEY
bedrock	us.anthropic.claude-sonnet-4-20250514-v1:0	Uses AWS credentials
copilot	claude-sonnet-4	Set base_url to Copilot endpoint
local	qwen3-8b	LM Studio, Ollama, or any OpenAI-compatible server

MCP servers

Two modes — pick what fits your setup:

With Amazon Q (default): just set server names. MergeScope finds them in ~/.aws/amazonq/mcp.json or .amazonq/mcp.json.

Standalone (no Amazon Q): define command, args, and env directly:

mcp:
  servers:
    github:
      name: "github-mcp-server"
      command: "npx"
      args: ["-y", "@anthropic/github-mcp-server"]
      env:
        GITHUB_TOKEN: "ghp_..."
    jira:
      name: "atlassian-mcp-server"
      command: "npx"
      args: ["-y", "@anthropic/atlassian-mcp-server"]
      env:
        JIRA_API_TOKEN: "..."
        JIRA_URL: "https://company.jira.com"

You can mix both — some servers inline, others from Amazon Q.

Environment variables

MERGESCOPE_REPO, MERGESCOPE_LLM_PROVIDER, MERGESCOPE_LLM_MODEL, MERGESCOPE_LLM_BASE_URL, MERGESCOPE_LLM_API_KEY, MERGESCOPE_JIRA_BASE_URL, MERGESCOPE_LANGUAGE

Requirements

• Python 3.11+
• GitHub and Jira MCP servers (standalone or via Amazon Q)
• An LLM: Anthropic API key, AWS credentials, GitHub Copilot, or a local server

Development

git clone https://github.com/martin5211/MergeScope.git
cd MergeScope
uv venv && uv pip install -e .
pytest