Claude MCP task management: synthesis on real data, not pasted spreadsheets

Claude shines at writing, summarization, and structured debate—but without context, even the best model in the room produces fiction. A PM exports a spreadsheet each morning, pastes it into Claude Desktop, and asks "What should we prioritize this week?" The answer reads beautifully and may match the live Kanban board only forty percent of the time.

Claude MCP task management closes that gap. Claude Desktop acts as an MCP client connected to WKFGo's server. Instead of paragraphs from memory, the model invokes list_tasks, decision_inbox, or executive_brief and builds synthesis on permission-scoped JSON—the same data the web app shows on /dashboard and /my-tasks.

The result: defensible meeting briefs, sprint narratives, and stakeholder drafts that trace back to task ids—not vibes.

Why Claude-only PM workflows stall without MCP

Teams that rely on copy-paste exports hit predictable walls:

• Copy-paste tax: Every question needs a fresh export; PMs become export operators, not managers.
• Cross-project blind spots: Claude does not know which project is active unless you state it every time.
• Executive summary drift: CEO slides reflect the PM's mood, not measured metrics from get_project_report.
• No permission boundary: Admin exports expose all finance rows; Claude should not see what the user cannot.
• Stale by arrival: The spreadsheet was accurate at 8 AM; by the 2 PM meeting, three cards moved.

MCP enforces FeatureAccess per user—the same rules as the web app. A forbidden response on finance_summary means fix the role, not switch to an admin key.

Six-step framework: Claude Desktop + WKFGo MCP

Step 1: Install current Claude Desktop

You need a build with MCP support (official macOS and Windows releases). Web chat at claude.ai does not expose MCP—Desktop or compatible clients are required.

Step 2: Issue a personal wk_ key

WKFGo → Settings → Integrations → Connect your AI IDE. Each PM and executive gets their own key for auditability. Never share one admin token across the team repo.

Step 3: Add MCP config in Claude

Settings → Developer → Edit Config—add the WKFGo server with an Authorization header carrying your wk_ key. Exact paths and examples live at /api-docs.

Step 4: Run a read-only smoke test

Ask Claude to call list_tasks for assignments in a known project. Verify names match /my-tasks in the browser. Then try search_docs for a wiki page you know exists.

Step 5: Establish leadership workflows

• Morning: my_day plus a prioritization narrative for async standup posts.
• Pre-board: executive_brief with role=ceo or role=cfo as appropriate.
• Scope decisions: Read the top decision_inbox item, then simulate_scenario before committing.

Step 6: Store prompt templates in wiki

Create a wiki page "Claude MCP Prompts" with five templates—consistent onboarding for new PMs. Refresh monthly; do not reinvent prompts every sprint.

Anti-patterns

• Pasting full finance exports instead of calling finance_summary—leak risk, staleness, and no permission audit trail.
• Calling executive_brief without specifying role—CTO lens differs sharply from CFO or CHRO.
• Ignoring decision_inbox while hand-building priority lists in chat—the inbox exists for structured choice.
• One shared Claude account with one shared wk_ key—no attribution when something goes wrong.
• Treating Claude output as investor-ready without spot-checking FACTS in the web app.

WKFGo tools for Claude workflows

WKFGo's in-app chatbot handles quick browser questions via ChatbotContainer; Claude MCP serves deep synthesis sessions at the desk. One PostgreSQL backend; many clients.

Claude-specific use cases worth running weekly

Product manager before steering: "Read decision_inbox, summarize the top item with evidence, propose three options with owners." Claude narrates; MCP supplies overdue counts and finance flags.

CFO lens: "executive_brief role=cfo for the portfolio—three decisions with timing." Compare burn signals in /finance with brief FACTS before the board pack goes out.

Sprint PM: "list_tasks in the current sprint grouped by status—build an exception list." Use it to prep a 15-minute standup instead of a 45-minute round-robin.

Knowledge retrieval: "search_docs for deploy runbook—pre-release checklist." Wiki vector search grounds answers in project knowledge, not generic model memory.

One-hour PM team workshop

0–15 min: MCP connect plus list_tasks smoke test.
15–30: Live demos: my_day, search_docs, decision_inbox.
30–45: Each PM writes one personal prompt template.
45–60: Q&A; publish wiki page "Claude MCP Team Prompts."

Outcome: every PM with their own wk_ key—not a centralized admin token in a shared config file.

Board meeting prompt template (store in wiki):

1. Read decision_inbox top item
2. executive_brief role=ceo for that project
3. If scope decision: compare simulate_scenario cut_scope vs delay_release
4. Output: FACTS (bullets), DECISION (3 items with owners), dataGaps (honest)

FAQ — Claude MCP task management

Claude.ai web vs Desktop?
MCP requires Desktop or compatible clients—web chat has no MCP tool access.

Persian prompts on English task data?
Yes—WKFGo UI is bilingual (FA/EN); Claude synthesizes in either language. Jalali due dates appear in calendar utilities.

Does it replace human PMs?
No—Claude drafts; humans approve releases, assignments, and stakeholder commitments.

How do we secure MCP access?
Personal wk_ keys per user; revoke immediately on offboarding. Keys belong in password managers, never in git.

Next step

Connect Claude Desktop to WKFGo MCP with your personal wk_ key. Run one smoke test. Then try the board meeting template on a real project before your next leadership sync.

• Sign up
• Features
• MCP docs
• Pricing
• Contact