User: Please generate a single-page resume from my experience

AI Agent:
1. 📝 Write the resume to resume.md
2. 🔍 Call render_resume_pdf with its path
3. ⚠️ Failed: overflows by 6% (about 4 body lines); the bullet on line 58 ends with a 6-character line
4. 🔧 Edit only those lines in resume.md
5. ✅ Success! PDF generated

# Enter MCP Server directory and install dependencies
cd mcp_server
pip install -r requirements.txt
# ⚡ Or using uv (recommended): uv pip install -r requirements.txt

# Install Chromium browser (first time only, ~150MB)
playwright install chromium

{
  "mcpServers": {
    "resume-onepage-autofit-mcp": {
      "command": "python",
      "args": ["<your-path>/myresumebuilder/mcp_server/mcp_server.py"]
    }
  }
}

User provides experience → AI generates Markdown resume
                                    ↓
                        MCP Server renders & validates
                                    ↓
                    ┌─────── Detect page height ───────┐
                    │                                  │
                  Success                           Failure
               (within one page)                 (overflow X%)
                    │                                  │
              Generate PDF                Return per-block line counts
                    │                                  ↓
                    │                       AI edits the lines it chooses
                    │                                  │
                    └──────────── Re-render ←──────────┘

You build a one-page resume with the render_resume_pdf tool.

1. Write the resume to a Markdown file, e.g. /home/you/resumes/acme/resume.md, in the
   canonical forms described in the tool's markdown_path parameter. Write the file once.
2. Call render_resume_pdf with the file's absolute path.
3. After each render, change only the lines that need it with your edit tool and render
   again with the same path. Never rewrite the whole file or paste the resume into the call.
   - Start with the paragraph under pdf_path: it says whether the render succeeded, why not,
     and what the next render must achieve.
   - overflow: "Page fit" gives overflow_percent and overflow_body_lines. The table "Space by
     section and entry" shows how many lines each section, job, project and school takes. Its
     "wrapped items" column lists items that wrap, as "source line: characters on last line";
     "58: 6" means 6 characters on line 58 cost a whole line, so trimming them saves it. Cut
     what is least relevant to the target job first.
   - success with a large empty_space_percent: there is room; add relevant detail or stop.
   - layout_error: each line under "Layout warnings" is an entry header that wrapped. It must
     fit on one line.
   - Format warnings: in the same edit, rewrite each listed line to its expected form. If a
     month is missing, ask the user; never invent one.
4. Stop when status is success and there are no format warnings. If it still does not fit
   after 5 renders, tell the user what you would cut and ask.

Never change facts (numbers, dates, company names, titles) to make the page fit.

# Use VS Code Live Server extension (recommended)
# Right-click control_panel.html -> "Open with Live Server"

# Or Python simple server
python -m http.server 8080
# Visit http://localhost:8080/control_panel.html

# Clone repository
git clone https://github.com/seriserendipia/resume-onepage-autofit-mcp.git
cd resume-onepage-autofit-mcp

# Create virtual environment
conda create -n agent_env python=3.10
conda activate agent_env

# Install dependencies
pip install -r mcp_server/requirements.txt
playwright install chromium