Humane Proxy

Name: Humane Proxy
Author: vishisht16

by vishisht16communityv0.3.1

HumaneProxy is a lightweight AI safety middleware that intercepts user messages to detect self-harm ideation or criminal intent, alerting operators and providing empathetic responses before messages reach the LLM.

From the registry: AI safety middleware — detects self-harm and criminal intent in LLM prompts.

GitHub Repository

Install & configure

Prerequisites

required
Install humane-proxy[mcp]
Install the Python package with the MCP extra
```
$ pip install humane-proxy[mcp]
```

Install in your client

Paste into the AI you're chatting with — it will figure out the rest

Please install the `humane-proxy` MCP server into my current AI client (that's you).

Required prerequisites (do these first if not already done):
- **Install humane-proxy[mcp]** — Install the Python package with the MCP extra Run: `pip install humane-proxy[mcp]`

Canonical MCP server config (stdio transport):
- command: `humane-proxy`
- args: ["mcp-serve"]
- optional environment variables:
  - `HUMANE_PROXY_ENABLED_STAGES`: Active safety pipeline stages (e.g., '1,2,3') (example: `1`)
  - `HUMANE_PROXY_STAGE3_PROVIDER`: Stage 3 reasoning LLM provider (auto|openai_moderation|llamaguard|openai_chat|none) (example: `auto`)
  - `OPENAI_API_KEY`: OpenAI key for Stage 3 moderation (optional) (example: `<your-openai-api-key>`)
  - `GROQ_API_KEY`: Groq key for LlamaGuard in Stage 3 (optional) (example: `<your-groq-api-key>`)
  - `HUMANE_PROXY_STORAGE_BACKEND`: Storage backend (sqlite|redis|postgres) (example: `sqlite`)

Note: MCP server mode exposes check_message_safety, get_session_risk, list_recent_escalations. Stage 1 heuristics work with no API keys. Also supports HTTP transport via 'mcp-serve --transport http'.

Add this MCP server to my current client's config in the correct format for you. If you need secrets or credentials I haven't provided, ASK me — do not invent values or leave raw placeholders. After adding it, tell me how to verify the server is connected.

Note: MCP server mode exposes check_message_safety, get_session_risk, list_recent_escalations. Stage 1 heuristics work with no API keys. Also supports HTTP transport via 'mcp-serve --transport http'.