This setup is for VS Code / Cursor / Claude Code / Antigravity / OpenCode when you want:
- no Anthropic/OpenAI API keys,
- no local model runtime,
- your currently selected chat model to do the reasoning.
The original full Shannon pipeline runs its own model backend, so it cannot directly inherit Cursor’s selected chat model from inside an external process.
This MCP setup gives your AI client Shannon tools (save_deliverable, generate_totp) and lets your active chat model drive the security workflow.
- Clone:
git clone https://github.com/Tide-Trends/shannon.git
cd shannon- Make launcher executable:
chmod +x ./scripts/run-shannon-helper-mcp.sh- In your MCP config, set:
command: /absolute/path/to/shannon/scripts/run-shannon-helper-mcp.sh
env: SHANNON_TARGET_DIR=/absolute/path/to/target-project
Ready templates are in mcp-configs/.
Use my CURRENT selected chat model. Do not call any external model APIs.
Use Shannon MCP tools only:
- save_deliverable
- generate_totp
Target project path: /absolute/path/to/target-project
URL under test: https://example.com
Perform a defensive security assessment (authorized target only):
1) Build recon notes and save with save_deliverable (RECON)
2) Analyze top risks (auth/authz/xss/injection/ssrf)
3) Save findings as Shannon deliverables via save_deliverable
4) Provide remediation checklist with severity and fix priority
Use my currently selected model for all reasoning.
- Templates now point to
run-shannon-helper-mcp.sh(chat-model-native mode) - No local-model env vars required
- No cloud API keys required for this MCP flow