Talent Augmenting Layer MCP Server

Talent-Augmenting Layer — Personalised AI Augmentation Layer

Works with: ChatGPT | Claude | Gemini | Cursor | Windsurf | Any LLM

Make workers better, not dependent. A personalised AI augmentation system that follows you across every platform.

---

What Is This?

Talent-Augmenting Layer (TAL) is a personalised AI augmentation layer that transforms how AI interacts with you. It works with any LLM, on any platform, through a 4-tier architecture designed for cross-platform portability. Instead of treating you as a generic user, TAL:

1. Assesses your expertise, goals, work style, and growth areas using the TALQ (Talent-Augmenting Layer Questionnaire)
2. Creates a living profile in portable markdown that calibrates all AI interactions to YOUR context
3. Adapts — coaching you in areas where you're growing, accelerating you where you're expert, automating what should be automated, and protecting skills at risk of atrophying
4. Evolves — your profile updates as you grow, keeping the AI aligned with your changing needs
5. Travels with you — the same profile works in ChatGPT, Claude, Gemini, Cursor, Windsurf, or any LLM

The core insight: AI that does everything for you makes you worse over time. AI that knows WHEN to help, WHEN to coach, WHEN to challenge, and WHEN to step back makes you permanently better.

---

The Problem

Current AI tools have one mode: maximum helpfulness. This creates three failure patterns:

Talent-Augmenting Layer exists to prevent all three.

---

How It Works

Architecture: 4 Tiers

Talent-Augmenting Layer is a layer, not a product tied to one platform. It works through 4 tiers, from zero-dependency copy-paste to a full hosted web app:

┌─────────────────────────────────────────────────────────────────┐
│  Tier 4: Hosted Web App                                         │
│  Browser-based · Google OAuth · LLM assessment · email check-ins│
├─────────────────────────────────────────────────────────────────┤
│  Tier 3: MCP Server                                             │
│  14 tools · 5 resources · 4 prompts · automatic tracking        │
│  Claude Code · Cursor · Windsurf · Claude Desktop               │
├─────────────────────────────────────────────────────────────────┤
│  Tier 2: Platform-Native                                        │
│  Custom GPTs · Gemini Gems · Claude Projects                    │
│  Persistent context · conversation starters                     │
├─────────────────────────────────────────────────────────────────┤
│  Tier 1: Universal System Prompt                                │
│  Any LLM · zero dependencies · copy-paste setup                 │
└─────────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────────┐
│                  profiles/pro-{name}.md                          │
│  Portable markdown · same format across all tiers               │
│  Identity · Expertise Map · TALQ Scores · Task Classification   │
│  Growth Trajectory · Contrast Libraries · Red Lines             │
└─────────────────────────────────────────────────────────────────┘

All tiers share: same TALQ instrument, same scoring,
same profile format, same behavioural rules.

The Five Modes

Every task gets classified into one of five AI interaction modes:

Research-Backed Techniques

---

Quick Start

New to Claude Code and TAL MCP? Start with the first-time guide: docs/CLAUDE_CODE_FIRST_TIME_SETUP.md.

Pick the option that matches your setup:

Any LLM (2 min)

1. Paste universal-prompt/ASSESSMENT_PROMPT.md into a conversation. Answer the questions. Save the generated profile.
2. Paste universal-prompt/SYSTEM_PROMPT.md + your profile into your LLM's custom instructions.
3. Done. The AI now adapts to your expertise, coaches your growth areas, and protects your skills.

Custom GPT / Gem / Project (5 min)

Pre-configured instances with persistent context and conversation starters:

• ChatGPT: Import platform-configs/chatgpt-gpt.json as a Custom GPT
• Gemini: Follow platform-configs/gemini-gem.md to create a Gem
• Claude: Follow platform-configs/claude-project.md to set up a Project

MCP Server (10 min)

Full tool integration with automatic tracking:

cd mcp-server && pip install -e .

Add to your MCP client config (Claude Code, Cursor, Windsurf, Claude Desktop):

{
  "mcpServers": {
    "talent-augmenting-layer": {
      "command": "python",
      "args": ["-m", "src.server"],
      "cwd": "/path/to/talent-augmenting-layer/mcp-server",
      "env": {
        "TALENT_AUGMENTING_LAYER_PROFILES_DIR": "/path/to/talent-augmenting-layer/profiles"
      }
    }
  }
}

Run talent-assess as an MCP prompt to create your profile. If you want the Claude Code slash command /talent-assess, open this repository in Claude Code so it loads .claude/commands/, or copy those command files into ~/.claude/commands/.

Hosted Web App

Browser-based app with Google login, LLM-powered assessment, and email check-in reminders:

cd hosted && docker build -t talent-augmenting-layer . && docker run -p 5000:5000 --env-file .env talent-augmenting-layer

See hosted/README.md for full setup (OAuth credentials, LLM API key, SMTP config).

Day-to-Day Commands (MCP / Claude Code)

• /talent-assess — Run initial assessment or full re-assessment
• /talent-update — Update profile based on recent interactions
• /talent-coach — Start a targeted coaching session on a specific skill

These slash commands are separate from the MCP server prompts. The MCP server exposes talent-assess, talent-coach, and talent-update as prompts. In MCP usage, the conversation is powered by your selected client model (for example, your Claude Code model), while the server provides tools and profile storage.

See docs/integration-guide.md for detailed platform-specific instructions.

---

Cross-Platform Portability

Talent-Augmenting Layer is designed as a layer -- not tied to any specific tool, LLM, or platform. The 4-tier architecture means it works everywhere:

The profile is portable markdown -- it works anywhere you can inject system context. Take your profile from Claude Code to ChatGPT to Cursor and back. Your AI calibration follows you.

---

Research Foundation

Built on empirical research, not opinions:

---

File Structure

talent-augmenting-layer/
├── CLAUDE.md                           # Core system prompt (the brain)
├── README.md                           # This file
├── CITATION.cff                        # Machine-readable citation metadata
├── LICENSE                             # CC BY-NC-SA 4.0
├── COPYRIGHT                           # Attribution notice
├── .claude/
│   ├── commands/
│   │   ├── talent-assess.md         # /talent-assess slash command
│   │   ├── talent-update.md         # /talent-update slash command
│   │   └── talent-coach.md          # /talent-coach slash command
│   └── settings.local.json            # Claude Code permissions
├── universal-prompt/                   # Tier 1: Works with any LLM
│   ├── SYSTEM_PROMPT.md                # Full system prompt (~4k tokens)
│   ├── SYSTEM_PROMPT_COMPACT.md        # Compact version for token-limited platforms
│   ├── ASSESSMENT_PROMPT.md            # Self-contained assessment prompt
│   └── QUICK_START.md                  # Step-by-step setup instructions
├── platform-configs/                   # Tier 2: Pre-configured platform instances
│   ├── chatgpt-gpt.json               # ChatGPT Custom GPT configuration
│   ├── gemini-gem.md                   # Gemini Gem setup guide
│   └── claude-project.md              # Claude Project setup guide
├── mcp-server/                         # Tier 3: Cross-platform MCP server
│   ├── pyproject.toml                  # Package config
│   ├── README.md                       # MCP server docs
│   └── src/
│       ├── server.py                   # MCP tools, resources, prompts (14 tools)
│       ├── profile_manager.py          # Profile CRUD, parsing, interaction logging
│       └── assessment.py               # Embedded assessment engine (questions, scoring)
├── hosted/                             # Tier 4: Standalone web application
│   ├── app.py                          # Flask application (routes, OAuth, assessment)
│   ├── config.py                       # Environment configuration
│   ├── database.py                     # Database models and persistence
│   ├── llm_client.py                   # LLM integration for conversational assessment
│   ├── scoring.py                      # TALQ scoring engine
│   ├── auth.py                         # Google OAuth authentication
│   ├── email_service.py                # 2-week check-in email reminders
│   ├── scheduler.py                    # Background task scheduling
│   ├── templates/                      # HTML templates (login, assessment, dashboard, checkin)
│   ├── static/                         # CSS and JavaScript
│   ├── requirements.txt                # Python dependencies
│   ├── Dockerfile                      # Container deployment
│   └── README.md                       # Hosted app setup guide
├── assessment/
│   ├── framework.md                    # Assessment methodology
│   ├── scoring-instrument.md           # TALQ psychometric instrument
│   ├── coaching-modules.md             # Structured coaching sessions (5 modules, 13 sessions)
│   ├── ab-testing-framework.md         # A/B testing design for outcomes research
│   └── literature-foundations.md       # Research backing
├── dashboard/
│   └── app.py                          # Streamlit org-level analytics dashboard
├── web-ui/
│   └── index.html                      # Standalone web assessment UI
├── docs/
│   └── integration-guide.md            # 4-tier integration guide
├── profiles/
│   ├── TEMPLATE.md                     # Blank profile template
│   └── pro-angelo.md                   # Example: Angelo's profile
└── context/                            # Research papers (Buçinca, Acemoglu, Mollick)

Related project: Talent-Augmenting Layer Benchmark -- a 3-layer evaluation framework for measuring whether LLMs augment or replace human intelligence.

---

What Makes This Different From Memory?

Good question. Memory stores facts. Talent-Augmenting Layer is how memory is used.

Memory is the database. TAL is the operating system.

---

Contributing

This is an open-source personalised AI augmentation layer. Current status:

• Core system prompt with 4 interaction modes (CLAUDE.md)
• Interactive assessment with profile generation (TALQ)
• Psychometric scoring instrument with validated Likert scales
• Tier 1: Universal system prompt for any LLM (4 files)
• Tier 2: Platform-native configs for ChatGPT, Gemini, Claude (3 files)
• Tier 3: MCP server with 14 tools, 5 resources, 4 prompts
• Tier 4: Hosted web app with Google OAuth, LLM assessment, email check-ins
• Embedded chatbot-driven onboarding (any MCP client can run the assessment)
• Organisation-level dashboard (Streamlit)
• Skill progression tracking with trend analysis and atrophy detection
• 4-tier integration guide with cross-platform sync
• A/B testing framework for outcomes research
• HTTP transport for enterprise MCP deployment
• Integration with existing L&D platforms
• Multi-user benchmarking and anonymized comparisons
• API middleware for any LLM provider
• Mobile-friendly assessment UI

---

License

This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

You are free to share and adapt this work for non-commercial purposes, as long as you give appropriate credit and distribute contributions under the same license.

See LICENSE for the full text.

---

Citation

If you use Talent-Augmenting Layer in research or publications, please cite:

@software{leone2026talentaugmentinglayer,
  author    = {Leone, Angelo},
  title     = {Talent-Augmenting Layer: A Personalised AI Augmentation Layer for Workforce Development},
  version   = {0.2.0},
  year      = {2026},
  url       = {https://github.com/angelo-leone/talent-augmenting-layer},
  license   = {CC-BY-NC-SA-4.0}
}

Or see CITATION.cff for machine-readable citation metadata.

---

Built by Angelo Leone at PUBLIC. Powered by research from Buçinca, Acemoglu, Mollick, Drago & Laine. Every interaction should leave you more capable, not more dependent.

Copyright (c) 2026 Angelo Leone. All rights reserved under CC BY-NC-SA 4.0.