Agile Project Lifecycle + MCP (Replit-Style, Full Loop + HITL)

Same phases and gates as the base agile-project-lifecycle, with MCP (Model Context Protocol) used for live docs and API schemas during discovery, planning, and development. Invoke this skill when the user wants the agile lifecycle and MCP for docs/APIs.

When to use (MCP variant)

The user has a rough idea that involves external APIs or libraries (e.g. Stripe, React, Prisma) and wants current docs/schemas via MCP instead of pasting docs.
You want the full agile flow (requirements → develop → unit → integration → e2e → feedback loop + HITL) with MCP at discovery and develop.
MCP server setup: Use a configured MCP server (e.g. project-docs, Context7, Stripe). In this repo: examples/mcp-servers/README.md and .cursor/mcp.json.example. Run npm install in examples/mcp-servers/project-docs/, copy the example to .cursor/mcp.json, restart Cursor.

MCP best practice: check tool schema first

Before calling any MCP tool, read the tool’s descriptor (e.g. mcps/<server>/tools/<tool>.json) so you use correct parameters and required fields.

Phase 0: Discovery and requirements

Research: Use explore (or main agent) to understand the codebase. Use MCP to fetch current docs or API schema for the idea (e.g. Context7 for library docs, project-docs for repo docs, Stripe MCP for payments). Use this to clarify scope and write accurate requirements. Check each MCP tool’s schema before calling.
Clarify: Summarize what "done" looks like in 2–5 sentences. If the idea is vague, ask 1–2 short clarifying questions and incorporate answers.
Requirements: Produce a short requirements section the human can approve (user-facing outcomes, acceptance criteria, out of scope, technical constraints).

Then output Gate 1: Requirements (same as base skill). Stop. Do not start development until the user has approved.

Phase 1: Iteration planning

Invoke the planner subagent with the approved requirements. Optionally use MCP to refine scope (e.g. "what endpoints exist?" from an API schema). Get slices and current slice scope/steps. Present the iteration plan.

Phase 2: Development cycle (loop)

Same structure as base skill (2.1 Develop → 2.2 Unit → 2.3 Integration → 2.4 E2E → 2.5 Verify → 2.6 Gate 2). MCP-specific:

2.1 Develop (with MCP)

Implement the current slice. Call MCP whenever you need up-to-date API or library details (signatures, options, examples). Check tool schema before each MCP call. Follow project rules. Keep the project runnable.

2.5 Verify (optional MCP)

Verifier can use MCP to check implementation against current docs (e.g. "does this match the official API?").

2.6 Feedback gate (Gate 2)

Same as base skill. Optional: Use beforeMCPExecution / afterMCPExecution hooks to audit or gate MCP usage; see .cursor/hooks/examples/mcp-hooks.example.json and .cursor/rules/mcp-patterns.mdc.

Phase 3: Completion

Same as base skill. Run only when the user said "Complete" or approved the final iteration.

Loop summary

Same as base skill. Never assume approval. For test levels, see reference.md. For MCP server setup and hooks, see examples/agent-frameworks/agile-project-lifecycle-mcp.md and .cursor/rules/mcp-patterns.mdc.