MCP-1st-Birthday/team-synapse

Team Synapse – MCP in Action (Track 2)

An agentic Gradio app that analyzes sales meetings, writes to a Neo4j knowledge graph, and uses MCP tools (Neo4j + Miro) to reason, plan, and execute.

• —Track tag: "mcp-in-action-track-enterprise"
• —Bonus areas: Context engineering (agent tool reasoning + knowledge graph), visual workflow (Miro), RAG-like graph querying

🎬 Watch Demo Video

🔗 Connect


Quick Start (Local)

Prereqs: Python 3.10+, Neo4j 5.x, Miro API token and a board.

1) Install dependencies

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

2) Set environment

# Tenant / demo
export TEAM_SYNAPSE_TENANT_ID=demo

# Neo4j
export NEO4J_URI=bolt://localhost:7687
export NEO4J_USER=neo4j
export NEO4J_PASSWORD=your_password
export NEO4J_DATABASE=neo4j

# Gemini / Google
export GOOGLE_API_KEY=your_key

# Miro (optional but recommended for Track 2 visual action)
export MIRO_API_TOKEN=your_miro_token
export MIRO_BOARD_ID=your_board_id

# Use real MCP server as tools (Track 2 requirement)
export USE_MCP_SUBPROCESS=true

3) Run the app

python3 app.py

Open the Gradio URL and upload a sample audio file. The app will:

• —Transcribe + analyze with Gemini
• —Store insights in Neo4j
• —Run the MEDDPICC agent that uses MCP tools to query the graph and create visual notes in Miro

MCP Tools

We provide a real FastMCP server (mcpserver.py) and an in-process DirectToolExecutor fallback for Spaces. To meet Track 2 (“use MCP servers as tools”), set `USEMCP_SUBPROCESS=true`.

• —query_graph → Neo4j read (MATCH-only security)
• —add_node → Create a graph node AND (optionally) auto-link to a client/meeting via relationship
• —link_node → Attach any existing node to a client/meeting
• —findmissingroles → MEDDPICC gap detection (Kill Shot)
• —postriskto_miro → Create a sticky note (supports annotations: layoutSlot, clientName, meetingId, nodeId)
• —read_board → List board items
• —connect_items → Draw connector between two Miro items

Testing

• —Quick test: python3 test_mcp_tools.py (skips write ops by default; reads board if MIRO_* env vars set)
• —Manual Miro test:

python -q <<'PY'
from mcp_tools import create_sticky_note
print(create_sticky_note("Track 2 test", "red", 0, 0))
PY

Deploy to Hugging Face Spaces

1. Create Space

Go to huggingface.co/new-space → SDK: Gradio

2. Push code

git remote add hf https://huggingface.co/spaces/YOUR_USERNAME/team-synapse
git push hf main

3. Add Secrets (Settings → Repository secrets)

Note: For Google credentials, paste the entire JSON file content into GOOGLE_APPLICATION_CREDENTIALS_JSON. The startup.py script handles writing it to a temp file.

HF Space Submission Checklist

• —[ ] Publish as a Space under the event org
• —[ ] README includes tag: "mcp-in-action-track-enterprise"
• —[ ] Demo video (1–5 min) showing: agent gap detection, graph queries, Miro actions
• —[ ] Social media post link in this README
• —[ ] Team members listed (HF usernames) if applicable

Notes on Context Engineering

• —Agent system prompts enforce “reason before tool calls” and limit tool loops
• —Graph-first orchestration: client-centric model, MEDDPICC elements, and explicit linking for new nodes

Troubleshooting

• —Neo4j “Collections containing null values…” → We sanitize lists before storage in services/neo4j_service.py
• —Miro not creating items → Verify MIROAPITOKEN and MIROBOARDID. Use get_board_items() to confirm access
• —Real MCP mode not connecting → Ensure USE_MCP_SUBPROCESS=true and that Python can spawn subprocesses in your environment