saran.
Menu
Resume LensEvidence-first resume review

Resume Lens · version 0.2.1 · English v1

An ATS score you can inspect.

A resume can be rejected for missing or hard-to-read basics before a person sees it. Resume Lens reviews a PDF, DOCX or plain-text resume, shows each readiness deduction and suggests a fix. A job description is optional: add one when you want a separate keyword-match score and evidence gaps.

Each deduction links back to a specific check.

Download the source packageSee an example report

Release source archive · version 0.2.1 · SHA-256a3797884b980b299bb90a4bea9295504fa9e5d95ff071607ab55daeeaa29c39f

Resume Lens wordmark beside a document under an amber inspection lens.

Two ways to work

Review files locally or send text from your agent.

01 / LOCAL

Offline / local file mode

Download the package, choose narrow folders on your computer, and let the installed engine read PDF, DOCX, or TXT files. After setup, the scorer makes no network or model call.

  • Best when the source files should stay on your device.
  • Initial npm ci --ignore-scripts downloads pinned libraries; later scoring runs locally.
  • Scanned PDFs still need OCR outside this version.

02 / HOSTED

Hosted text mode

The hosted edition works with compatible MCP clients. People choose to send resume text and an optional job description through their agent; it uses the same rule-based review, with no LLM, account, or saved resume.

  • It accepts text you deliberately provide; this static website has no upload form.
  • The service is designed for per-request memory and no application-body logging.
  • Hosted requests are checked against free evaluation limits. Oversized input returns an explicit error instead of being shortened.
  • For longer documents and repeated reviews, use local file mode.

What the number means

One resume score. A separate job match.

ATS readiness checks readable text, contact patterns, populated sections, dates and writing signals. Each check shows its points and a useful fix. PDF/DOCX inspection includes supported layout risks; pasted text leaves formatting unassessed.

Job match measures weighted keywords in the supplied role. Evidence coverage is stricter: each requirement is checked against what the resume actually says.

How the review moves

Start with a score. Go deeper when useful.

For a score, inspect the resume and score it. The remaining tools support optional job matching, deeper review, comparison and reports.

  1. inspect_resume

    Extract PDF, DOCX, or plain-text content and show warnings before you rely on it.

  2. prepare_requirements

    Separate what a job requires from what it prefers, then flag unclear requirements for review.

  3. review_evidence

    Correct the extracted text or job requirements and save the evidence you have reviewed.

  4. score_resume

    Get ATS readiness out of 100 and prioritized fixes, with no JD required. Optional job matching stays separate.

  5. compare_versions

    Compare two versions only when their job description, rubric, as-of date, and denominator still match.

  6. export_report

    Create a new local HTML or JSON report without overwriting an earlier export.

Local setup

Start in the folder you downloaded.

  1. Install Node.js 24 or later, then unzip the source package.
  2. Open a terminal inside the extracted resume-lens-mcp-0.2.1 folder.
  3. Run the three commands below to build the tool and create example reports.
npm ci --ignore-scripts
npm run build
npm run demo

A better review

Keep the original beside the score.

English v1 uses finite rules and a limited alias set. New phrasing, mixed-language content, unusual layouts, dates, negation, scope, and copied requirements can need a human decision.

PDF reading order can differ from what you see on the page. OCR, encrypted files, and malformed files are outside the supported path. Review extraction and the job-description model before relying on a result.

Local processing does not turn a cloud-backed agent conversation into a private offline conversation. Requesting full evidence or text can place excerpts in the harness's model context.