Skip to content
Workflow MCP

Workflow MCP

Never Lose Context Again

Structured workflows that track every decision, finding, and task

Get Started
claude — ~/my-project

Generated session files:

📄 RESEARCH_AUTH-SYSTEM_2026-01-21.md

Research Session: AUTH-SYSTEM

Created: 2026-01-21 14:32 | Status: In Progress | DPS: 4

Research Log

14:32 Analysis: Common AWS Auth Mechanisms

Entry-ID: RES-AUTH-001

Question: What auth mechanisms are commonly used with AWS?

MechanismUse Case
CognitoUser pools, federated identity
IAMService-to-service auth
STSTemporary credentials

14:35 Finding: Current Implementation Gap

Entry-ID: RES-AUTH-002

Evidence: src/auth/provider.ts:42

No refresh token rotation implemented. Using short-lived tokens only.

14:38 Decision: Recommend Cognito Integration

Entry-ID: RES-AUTH-003

Rationale: Provides built-in token refresh, MFA support, and AWS service integration.

📄 PLAN_AUTH-SYSTEM_2026-01-21.md

Implementation Plan: AUTH-SYSTEM

Created: 2026-01-21 14:45 | Status: In Progress | DPS: 1

Planning Log

14:45 Requirement: Cognito User Pool Setup

Entry-ID: PLN-AUTH-001

Source: RES-AUTH-003

Acceptance: User pool created with email/password and Google OAuth configured.

14:48 Requirement: Token Refresh Flow

Entry-ID: PLN-AUTH-002

Source: RES-AUTH-002

Acceptance: Refresh tokens stored securely, automatic refresh on 401.

14:52 Task: Update Auth Provider

Entry-ID: PLN-AUTH-003

Status: pending

File: src/auth/provider.ts

Replace current JWT logic with Cognito SDK integration.

Click anywhere to close

Features

Section titled “Features”

Structured Sessions

Start, pause, and resume workflow sessions with automatic Entry-ID generation, state persistence, and iteration tracking.

Dynamic Problem Solving

Control how strictly Claude follows your workflow with DPS levels (0-5) - from strict adherence to maximum flexibility.

Seven Workflow Phases

Built-in support for /research, /plan, /implement, /review, /debug, /refactor, and /archive with automatic input file linking.

Local LLM Delegation

Offload code generation to local models like DeepSeek, keeping Claude’s context clear for reasoning tasks.

Quick Start

Section titled “Quick Start”
Terminal window
# Start a research session
/workflow:research my-feature


# Plan based on research
/workflow:plan my-feature


# Implement the plan
/workflow:implement my-feature dps=2


# With local LLM enabled (requires Ollama)
/workflow:implement my-feature llm=true

Why Workflow MCP?

Section titled “Why Workflow MCP?”

AI assistants are powerful, but unstructured conversations can lead to:

  • Lost context and forgotten decisions
  • Inconsistent approaches across sessions
  • Difficulty tracking what was tried and why
  • Wasted tokens on mechanical code generation

Workflow MCP brings methodology to AI-assisted development:

  1. Structure - Every session follows a defined workflow phase with automatic input file linking
  2. Traceability - All findings, decisions, and blockers are logged with Entry-IDs
  3. Continuity - Sessions can be paused and resumed; iteration tracking detects repeated attempts
  4. Flexibility - DPS levels control how strictly Claude follows the plan
  5. Efficiency - Delegate mechanical coding to local LLMs, preserve context for reasoning