PM Context MCP Server

A Python MCP server that gives Claude access to PM workflow data (sprints, roadmap, blockers, workload) so you can ask it questions a PM actually needs answered.

No API keys required. Runs entirely on synthetic fixture data representing a fictional product team ("Petal & Co"). Designed to be forked and connected to a real Linear or Jira workspace privately.

<img src="Demo-mcp.jpg" alt="Demo of Claude answering a sprint question using the MCP server" width="50%">

---

The Problem

PMs spend a disproportionate amount of time context-switching: checking Linear for sprint status, Confluence for the roadmap, Slack for who's blocked. Each lookup is a tab, a search, a few seconds of reorientation.

The bigger problem is that the questions PMs ask are aggregate and cross-cutting ("what should I focus on today?", "who's overloaded?", "what's actually blocking us?") but the tools expose raw CRUD. You have to do the aggregation yourself.

This MCP server exposes PM-oriented tools to Claude so those questions get answered in one place, with actual data behind them.

---

Architecture

Claude (Desktop or CLI)
        │
        │  MCP (stdio transport)
        ▼
pm-mcp-server (FastMCP)
        │
        │  loads
        ▼
data/fixture.json          ← synthetic Petal & Co dataset
(or real Linear / Jira API in a private fork)

The server runs as a local subprocess. Claude calls tools over stdio; no network, no auth for the demo.

---

Tools

Tool	What Claude can ask
get_current_sprint_tool	"What's in the current sprint?" / "How close are we to done?"
get_my_issues_tool	"What's on Priya's plate right now?"
get_blockers_tool	"What's blocking the team?" / "What should we unblock first?"
search_issues_tool	"Find all issues related to payments"
get_roadmap_tool	"How far along are we on each epic?"
get_team_workload_tool	"Who has the most issues assigned?" / "Show me the Growth team's workload"
get_velocity_tool	"What's our sprint velocity trend?"

---

How to Run

Requirements: Python 3.10+, uv

# Clone and enter the repo
git clone https://github.com/jackhendon/pm-mcp-server
cd pm-mcp-server

# Install dependencies
uv sync

# Test a tool directly
uv run python -c "from tools.sprints import get_current_sprint; import json; print(json.dumps(get_current_sprint(), indent=2))"

# Run the MCP inspector (requires mcp[cli])
uv run mcp dev server.py

Connect to Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "pm-context": {
      "command": "/Users/yourname/.local/bin/uv",
      "args": ["--directory", "/path/to/pm-mcp-server", "run", "python", "server.py"]
    }
  }
}

Note: Claude Desktop on macOS doesn't inherit your shell PATH, so uv must be an absolute path. Run which uv in your terminal to find it.

Config location:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop. You should see pm-context listed under connected tools.

Connect to Claude Code (CLI)

claude mcp add pm-context uv run python server.py --cwd /path/to/pm-mcp-server

Try these prompts

• "What's in the current sprint and how close are we to finishing?"
• "Who has the most work on their plate right now?"
• "What's blocking the team and what should we unblock first?"
• "How has our sprint velocity trended over the last few sprints?"
• "What's the roadmap looking like across all epics?"
• "Find all issues related to API authentication"

---

Fixture Data: Petal & Co

The synthetic dataset represents a fictional gift-card startup's product team:

• 6 team members across 2 teams (Growth, Core)
• 3 projects: Checkout Redesign, API Platform v2, Gifting Suite
• 3 epics at different stages of completion
• 17 issues in the active sprint (Sprint 14) with realistic statuses and blockers
• 4 completed sprints for velocity calculation

The data is rich enough that all 7 tools return genuinely interesting and differentiated results.

---

Real API Mode

This repo is designed to be forked privately and connected to a real Linear or Jira workspace.

1. Fork the repo
2. Copy .env.example to .env and add your credentials
3. Implement tools/integrations/linear.py or tools/integrations/jira.py that returns data in the same shape as the fixture
4. Update tools/__init__.py to route to the real integration based on DATA_SOURCE env var

The tool signatures and return shapes stay the same. Claude doesn't know or care whether the data comes from a fixture or a live API.