ResumeTailor

MCP-Resume-Tailoris a Model Context Protocol server system that automates document editing through LibreOffice's UNO API. Two MCP servers — adocument editorand ajob-description keyword extractor— connect to Claude Desktop, enabling AI-driven resume customization from natural conversation.

The editor preserves full document fidelity by operating directly on.odtfiles through LibreOffice's internal document-object model rather than converting to intermediate formats. Bookmark-delimited sections define the editable regions, and content is replaced while retaining the template's formatting, styles, and layout. The keyword extractor analyzes job descriptions through the Claude API and returns structured, 5-tier ranked output that guides the tailoring process.

Goal:Preserve the template's full formatting, styles, and layout across every edit.

Rather than converting documents to Markdown or HTML — a lossy transformation — the editor connects to LibreOffice's UNO bridge over a local socket and operates on the live document model. Bookmark pairs delimit editable regions. Content replacement happens at the document-object level, so fonts, paragraph styles, spacing, and page structure are preserved exactly as the template author designed them.

Goal:Expose document editing and keyword extraction as first-class MCP tools, not REST wrappers.

Both servers are built directly on FastMCP with typed tool schemas. Claude Desktop invokes them through the Model Context Protocol — no glue code, no prompt injection, no API choreography. The AI decides when and how to call each tool based on the conversation context, combining keyword extraction output with document editing in a single interactive session.

Goal:Run locally for development, containerized for reliability — same MCP tool surface either way.

Local mode uses stdio transport: Claude Desktop launches the Python process directly. Docker mode runs both servers behind Nginx, with proxy scripts bridging Claude Desktop's stdio protocol to HTTP/SSE endpoints inside the container. The MCP tool surface is identical in both modes — no code changes, no feature differences.

Claude Desktop ──stdio──▶ resume_editor_server.py ──UNO──▶ LibreOffice (port 2002) ──▶ .odt Claude Desktop ──stdio──▶ jd_keyword_extractor_mcp.py ──API──▶ Claude API

Claude Desktop launches each MCP server as a child process. The document editor connects to a headless LibreOffice instance via UNO bridge on port 2002. The keyword extractor calls the Anthropic API directly.

Claude Desktop ──stdio──▶ macOS proxy ──HTTP──▶ Docker (Nginx) ├─ :5001 → Resume Editor MCP (port 8000, streamable-http) ├─ :5002 → JD Extractor MCP (port 9000, streamable-http) └─ LibreOffice headless (port 2002, internal)

The macOS proxy scripts translate between Claude Desktop's stdio MCP protocol and the containerized HTTP/SSE endpoints, handling MCP session ID management and server-sent event stream parsing. Nginx routes traffic to the correct internal service.

The document editor operates on bookmark-pair-delimited regions. The current template exposes:

Theextract_jd_keywordstool returns structured JSON with:

- 5-tier ranked keywords— critical, high, medium, low, and nice-to-have
- Requirements summaryand role context
- Tools, technologies, and methodologiesextracted from the posting
- Core responsibilitiesand desired outcomes
- Role level, scope, and company context

Problem:LibreOffice's UNO API requires a running headless instance with an open socket on port 2002. Connection drops, document locks, and process state leaks are common failure modes in long-running sessions.

Solution:The editor manages the full connection lifecycle — establishing the UNO bridge, loading documents via URL protocol, handling open/close state, and writing to timestamped output files to avoid clobbering the template. The Docker deployment uses a dedicated startup sequence to guarantee LibreOffice is accepting connections before the MCP server begins handling requests.

2. MCP Transport Bridging (stdio to HTTP)

Problem:Claude Desktop only supports stdio-based MCP transport. Containerized services expose HTTP endpoints. These are fundamentally different protocol models with incompatible stream semantics.

Solution:Custom proxy scripts (mac_proxy_resume_editor.py,mac_proxy_jd_extractor.py) bridge the gap: they read JSON-RPC messages from stdin, forward them as HTTP requests to the container, parse the SSE response stream back into MCP messages, and manage session IDs across the connection lifecycle. Claude Desktop sees a local stdio server; the container sees standard HTTP clients.

3. Bookmark-Preserving Content Replacement

Problem:Replacing text between bookmarks in an.odtdocument must preserve the surrounding document structure — paragraph styles, character formatting, page layout — without corrupting the underlying XML.

Solution:The UNO API operates on LibreOffice's live document-object model, not raw XML.replace_bookmark_range_text()enumerates the text range between paired bookmarks (Section_Start/Section_End), removes existing content paragraph by paragraph, and inserts new content while inheriting the template's default paragraph and character styles.

- Python 3.10+
- LibreOffice with UNO support (libreoffice-script-provider-python)
- An.odtresume template with bookmark pairs (seeBookmark Setup Guide)
- For containerized deployment: Docker and Docker Compose
- For keyword extraction: Anthropic API key

git clone https://github.com/adi2355/MCP-Resume-Tailor.git cd MCP-Resume-Tailor python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt # Link system UNO modules into venv ln -s /usr/lib/python3/dist-packages/uno.py .venv/lib/python3./site-packages/uno.py ln -s /usr/lib/python3/dist-packages/unohelper.py .venv/lib/python3./site-packages/unohelper.py

soffice --accept="socket,host=localhost,port=2002;urp;" --headless --norestore --nologo --nodefault &

{ "mcpServers": { "LibreOfficeResumeEditor": { "command": "/path/to/MCP-Resume-Tailor/.venv/bin/python", "args": ["resume_editor_server.py"], "cwd": "/path/to/MCP-Resume-Tailor" }, "JDKeywordExtractor": { "command": "/path/to/MCP-Resume-Tailor/.venv/bin/python", "args": ["jd_keyword_extractor_mcp.py"], "cwd": "/path/to/MCP-Resume-Tailor" } } }

Configure Claude Desktop for containerized use:

{ "mcpServers": { "LibreOfficeResumeEditor": { "url": "http://localhost:5001" }, "JDKeywordExtractor": { "url": "http://localhost:5002" } } }

SeeDocker Deployment GuideandmacOS Setup Guidefor platform-specific details.

MCP-Resume-Tailor/ ├── resume_editor_server.py # MCP server: document section editing via UNO ├── jd_keyword_extractor_mcp.py # MCP server: JD keyword extraction via Claude API ├── edit_resume_uno.py # Core LibreOffice UNO API operations ├── container_api.py # Flask REST wrapper for container mode ├── mac_proxy_resume_editor.py # stdio → HTTP proxy (document editor) ├── mac_proxy_jd_extractor.py # stdio → HTTP proxy (keyword extractor) ├── docker-compose.yml # Multi-service container orchestration ├── Dockerfile # Ubuntu 22.04 + LibreOffice + Python ├── start_container_service.sh # Docker entrypoint script ├── start_resume_tailor_services.sh # Start both MCP services locally ├── requirements.txt # Python dependencies ├── claude_desktop_config.json # Claude Desktop config (stdio mode) ├── claude_desktop_config_docker.json # Claude Desktop config (HTTP/Docker mode) ├── CLAUDE_MCP_USAGE.md # Integration guide ├── DOCKER_README.md # Docker documentation ├── MACOS_SETUP.md # macOS setup guide ├── OPTIMIZED_SYSTEM_USAGE.md # Workflow documentation └── setup_resume_bookmarks.md # Template bookmark guide

An MCP server for reading and writing local XMind mind map files. Exposes over 25 tools that let any MCP-compatible AI client create, navigate, and edit .xmind files directly on disk.