AIP MCP Server

MCP server for the Applied Innovation Platform -- makes 14 AI innovation agents callable as tools from any Claude session, plus a Prototyper that generates working web apps from design briefs.

What It Does

• 14 agents on demand -- Call Scout, Empathy, Architect, Conductor, and 10 other agents without setting up a Claude Project
• Engagement lifecycle -- Create engagements, run pipeline stages, track DVFA scores, manage checkpoints
• Model-agnostic -- Auto-detects Claude, Gemini, GPT-4, or local models (via LiteLLM)
• Prototyper -- Generates React/Vite or static HTML prototypes from Visionary's design briefs

Quick Start

1. Prerequisites

• Python 3.10+ (install via uv python install 3.12 or Homebrew)
• The applied-innovation-platform repo cloned alongside this one

2. Install

git clone https://github.com/robertholzer42-stack/aip-mcp-server.git
cd aip-mcp-server
uv venv --python 3.12
source .venv/bin/activate
uv pip install mcp[cli] litellm pydantic anthropic uvicorn python-dotenv httpx

3. Configure

cp .env.example .env
# Edit .env: set your API key (ANTHROPIC_API_KEY, GOOGLE_API_KEY, etc.)
# Edit .env: set paths to your applied-innovation-platform directory

4. Register with Claude Code

Add to your project's .mcp.json or ~/.claude.json:

{
  "mcpServers": {
    "aip": {
      "command": "/path/to/aip-mcp-server/.venv/bin/python",
      "args": ["-m", "aip.server"],
      "cwd": "/path/to/aip-mcp-server/src",
      "env": {
        "AIP_SKILLS_DIR": "/path/to/applied-innovation-platform/skills",
        "AIP_ENGAGEMENTS_DIR": "/path/to/applied-innovation-platform/engagements"
      }
    }
  }
}

5. Use

From any Claude Code session:

"List the available innovation agents"
→ Calls list_agents tool

"Create a standard engagement analyzing AI in eldercare for healthcare executives"
→ Calls create_engagement tool

"Run the core analysis stage"
→ Calls run_stage tool

"Generate a prototype from the Visionary's design brief"
→ Calls generate_prototype tool

Tools (17)

Platform

Tool	Description
list_agents	Returns all 14 agents with codename, tier, role, frameworks
get_agent_skill	Returns the full SKILL.md content for any agent
get_platform_status	Agent count, engagement count, active LLM model

Engagement Lifecycle

Tool	Description
create_engagement	Creates an engagement with folder structure
run_agent	Runs a specific agent against an engagement
run_stage	Runs all agents for a pipeline stage
write_agent_output	Manually saves agent output to engagement folder
get_engagement_status	Shows completed agents, current tier, pending work

Pipeline

Tool	Description
run_full_pipeline	Orchestrates all 11 stages end-to-end
run_critic_review	Runs Critic quality review on a completed tier
get_dvfa_scores	Returns current DVFA scores with confidence and deltas

Checkpoints

Tool	Description
write_checkpoint	Saves engagement state for session continuity
read_checkpoint	Loads checkpoint data to resume in a new session

Prototyper

Tool	Description
generate_prototype	Generates a working web prototype from design briefs
list_prototypes	Lists all prototypes for an engagement
get_prototype_spec	Returns the spec used to generate a prototype
preview_prototype	Returns files and run instructions for a prototype

Model Support

The server auto-detects your LLM based on environment variables:

Priority	Env Variable	Model
1	AIP_MODEL	Explicit override (any LiteLLM model string)
2	ANTHROPIC_API_KEY	Claude Sonnet 4
3	GOOGLE_API_KEY	Gemini 2.0 Flash
4	OPENAI_API_KEY	GPT-4o
5	OLLAMA_BASE_URL	Llama 3 (local)

Architecture

src/aip/
  server.py          -- MCP server, 17 tool registrations
  config.py          -- Environment-based configuration
  models.py          -- Pydantic data models
  llm/
    router.py        -- Model-agnostic LLM routing
    claude.py        -- Anthropic SDK adapter
    openai_compat.py -- LiteLLM adapter (Gemini, GPT-4, Ollama)
  agents/
    registry.py      -- SKILL.md parser, agent catalog
    runner.py        -- Prompt construction, agent execution
  engagement/
    manager.py       -- Folder lifecycle, agent output I/O
    pipeline.py      -- 11-stage pipeline definitions
    checkpoint.py    -- Session continuity
    dvfa.py          -- DVFA score tracking
  prototyper/
    spec_builder.py  -- Extract specs from agent outputs
    generator.py     -- LLM-powered prototype generation
    templates/       -- React/Vite and static HTML scaffolds

License

Apache License 2.0