PyP6Xer MCP Server

Primavera P6 schedule analysis for AI assistants. Load an XER file, ask questions about critical path, float, schedule quality, earned value — get answers in seconds.

Works with Claude Desktop, ChatGPT, Claude.ai, or self-hosted with the included web UI.

Live server: https://pyp6xer-mcp.fly.dev/mcp (open, no key needed) PyPI: pip install pyp6xer-mcp Loading your own files | Authentication | Development

---

Try it now

The server has three sample projects pre-loaded. Connect and start asking questions — no upload needed.

Cache Key	Sector	Activities
cairo-alex-road	Roads	57
hilal-oil-gas	Oil & Gas	31
terminal-building-airport	Infrastructure	66

Example prompts:

• "Show the critical path for cairo-alex-road"
• "Run a schedule health check on terminal-building-airport"
• "Analyze float distribution for terminal-building-airport"

---

Setup

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "pyp6xer": {
      "command": "uvx",
      "args": ["pyp6xer-mcp"]
    }
  }
}

Runs locally via stdio. Load your own XER files by path: "Load /path/to/project.xer and analyze the critical path"

ChatGPT or Claude.ai

Add the remote server in your MCP settings:

https://pyp6xer-mcp.fly.dev/mcp

No API key needed. The sample projects above are ready to query. To analyze your own file, say "I need to upload a P6 schedule" — the AI will generate an upload link you can open in your browser.

Self-hosted (full stack with web UI)

git clone https://gitlab.com/articat1066/pyp6xer-mcp.git
cd pyp6xer-mcp
echo "OPENAI_API_KEY=sk-xxx" > .env
docker compose up

Open http://localhost:3000 — upload an XER file and start asking questions. The MCP server runs at http://localhost:5000 with sample projects auto-loaded.

Uses gpt-4o-mini by default. To change model or to use Claude, set ANTHROPIC_API_KEY and AI_MODEL=claude-sonnet-xxx in your .env.

MCP server only (no UI)

docker compose up pyp6xer-mcp

MCP endpoint at http://localhost:5000/mcp.

---

Loading your own XER files

Local file (Claude Desktop — file stays on your machine):

"Load /Users/me/Desktop/project.xer and show the critical path"

URL (any client):

"Load the XER file at https://example.com/project.xer"

Upload link (ChatGPT, Claude.ai — when you can't give a file path):

Ask the AI "I need to upload a schedule file." It calls the pyp6xer_get_upload_url tool, which returns a one-time upload link (valid 15 minutes). Open the link in your browser, upload your XER file, then come back to the chat. The file goes from your browser to the MCP server endpoint, not through the AI. It's held in memory for the session and lost on restart.

Direct HTTP upload (scripts, CI/CD):

curl -X POST -F "file=@project.xer" https://pyp6xer-mcp.fly.dev/upload
# Returns: {"cache_key": "upload_abc123", "activity_count": 250, ...}
# Use the cache_key with any analysis tool.

---

What it does

23 tools, 5 prompts, 3 resources. All tools are prefixed pyp6xer_.

Category	Tools
File Operations (5)	load_file, write_file, clear_cache, export_csv, get_upload_url
Project/Activity (5)	list_projects, list_activities, get_activity, search_activities, update_activity
Schedule Analysis (6)	critical_path, float_analysis, schedule_quality, relationship_analysis, slipping_activities, schedule_health_check
Resource Management (2)	list_resources, resource_utilization
Progress/Performance (4)	progress_summary, earned_value, work_package_summary, wbs_analysis
Structure (1)	list_calendars

See TOOLS.md for full parameter documentation.

---

Authentication

The hosted server at pyp6xer-mcp.fly.dev is open — no API key needed.

For self-hosted deployments, set API_KEY to enable authentication:

API_KEY=your_secret_key docker compose up

Clients then pass the key via query param, Bearer header, or x-api-key header:

{
  "mcpServers": {
    "pyp6xer": {
      "url": "https://your-server.com/mcp?api_key=YOUR_KEY"
    }
  }
}

---

Development

git clone https://gitlab.com/articat1066/pyp6xer-mcp.git
cd pyp6xer_mcp
uv pip install -e ".[dev]"

pytest                                      # 27 tests
python -m pyp6xer_mcp.cli                   # Run stdio (Claude Desktop)
python -m pyp6xer_mcp.cli streamable-http   # Run HTTP server on port 5000

black --line-length 100 .                   # Format
ruff check .                                # Lint

---

Architecture

pyp6xer_mcp/          # Python MCP server
├── server.py          # FastMCP instance
├── cli.py             # Entry point + sample auto-load
├── core/parser.py     # Shared XER parsing
├── tools/             # 23 MCP tools
├── prompts/           # 5 workflow prompts
├── resources/         # 3 MCP resources
└── http/              # Auth, upload, landing page

frontend/              # Next.js web UI (for self-hosted)
├── app/               # Pages + API routes (chat, tools)
├── components/        # Canvas, P6 upload, AI chat elements
└── Dockerfile

samples/               # Pre-loaded XER files (auto-load on startup)

Built on PyP6XER by Hassan Emam.

---

Built by

Paul Boucherat — portfolio · LinkedIn · GitLab · Buy me a coffee

If you're using this or have questions about P6 tooling, I'd like to hear from you.

Skills & Consulting

For AI-assisted P6 schedule analysis and consulting, see bouch.dev.

License

MIT — free for commercial and personal use.