QE Intelligence Suite

Structured QE analysis in Cursor, not in the browser

An open-source MCP server that standardizes risk-first, 11-section QE output — refinement, UAT, ticketless repo UAT, bug triage, and regression. Bring your own Anthropic key; optional Markdown artifacts under docs/qe-analysis/.

Problem and outcome

Teams skip refinement; prompt craft stays tribal. This encodes Senior QE judgment once.

The suite packages a sanitized system prompt (derived from a Cursor QE analysis skill) into five MCP tools. Each run produces the same section headings every time — risks, exploration-focused scenarios, mode-specific GO/NO-GO or regression guidance, and optional multi-repo ledger rules.

It is not the hosted QE assistant on this site (browser chat with a server-side API key). MCP runs locally over stdio in Cursor with your key and structured tool arguments.

Architecture

Cursor → stdio MCP → Anthropic → optional file write

Cursor (agent explores repo, fills tool args)
    │
    ▼
qe-refinement-mcp  ──►  Anthropic API  (BYOK in local .env)
    │
    └──►  docs/qe-analysis/*.md  (when save_file=true)

Five tools

One mode per tool — same output contract

ToolModeWhen to use
qe_refinementREFINEMENTBacklog and story refinement before development
qe_uatUATRelease readiness with a ticket or written acceptance criteria
qe_repo_uatREPO_UATTicketless UAT — explore the repo first, then pass feature area and hints
qe_bugBUGDefect triage and missed-coverage analysis
qe_regressionREGRESSIONBlast radius and retest strategy

Sample output

Sanitized fictional scope — full 11-section shape

Samples live in content/qe-intelligence-suite/samples/. See the GitHub repo for committed examples.

More in the repository.

Try it yourself

Local only — no API keys on this page

  1. Clone qe-intelligence-suite and run cd qe-refinement-mcp && npm install && npm run build.
  2. Copy .env.example .env and add your Anthropic key.
  3. Add the mcp.json snippet from the README (absolute paths to --env-file and dist/server.js).
  4. First smoke: call any tool with save_file: false before writing under docs/qe-analysis/.

Related demos on this site

  • QE showcase — strategy, pyramid, Promptfoo panel, ci-demo pyramid
  • QE assistant — browser chat with server-side Claude (different trust boundary)
  • CI dashboard — pipeline health demo / Supabase ingest

Showcase URL: https://arjunjhawar.dev/qe-intelligence-suite