Nefesh — Real Time Human State Awareness for AI

What Nefesh — Real-Time Human State Awareness for AI MCP server does

Nefesh — Real-Time Human State Awareness for AI MCP server gives an AI agent a structured view of a person’s changing physiological and behavioral state. The service accepts signals such as heart rate, heart-rate variability, voice tone, facial expression, and text sentiment, along with more than 30 supported fields. It returns a unified state that includes a stress score from 0 to 100 and a machine-readable suggested action.

Suggested actions indicate whether the agent should maintain its approach, simplify, de-escalate, or pause. On later calls within a session, the response can also report whether the previous action reduced stress. That feedback includes the prior score, current score, change in stress, and an effectiveness result, allowing an agent to evaluate its adaptation rather than reacting only to a single reading.

How it works

The service is hosted at https://mcp.nefesh.ai/mcp and uses Streamable HTTP, so clients do not need a local server installation. An agent can connect without a key, request one using the user’s real email address, and wait for the user to verify the emailed link. It then polls for activation and receives the key when the status becomes ready.

Authenticated calls use the X-Nefesh-Key header. The same service is also available through A2A at POST https://mcp.nefesh.ai/a2a, with the agent card published at /.well-known/agent-card.json. A2A uses the same authentication methods and exposes corresponding skills for state retrieval, signal ingestion, trigger memory, and session history.

The subject_id field on ingest supports trigger tracking for a subject. This lets later requests retrieve topics associated with active stress responses and topics that have been resolved over time.

Setup and configuration

Add the remote MCP URL to the client’s MCP configuration. A key can be supplied as an X-Nefesh-Key header, or the agent can guide the user through the email-verification flow. The free plan provides 1,000 calls per month, supports all signal types, allows 10 requests per minute, and does not require a credit card.

The README documents configuration locations for clients including Claude Desktop, Cursor, Windsurf, Claude Code, VS Code, Cline, Continue, Roo Code, JetBrains IDEs, Zed, OpenAI Codex CLI, and Gemini CLI. The exact configuration shape varies: VS Code uses servers, Zed uses context_servers, Codex uses TOML, and Continue uses YAML.

Nefesh — Real-Time Human State Awareness for AI MCP server requires the user to provide an accessible, real email address for self-provisioning. Disposable addresses are blocked, and the verification link must be clicked before the key becomes active.

Tools and capabilities

• request_api_key: Starts free-key provisioning after receiving the user’s real email address. It does not require authentication.
• check_api_key_status: Checks whether the emailed verification step is complete and returns either pending or ready with the key.
• get_human_state: Retrieves the current stress score, suggested action, and adaptation-effectiveness feedback.
• ingest: Sends biometric and behavioral signals and returns the resulting unified state. Include subject_id when trigger memory should be associated with a subject.
• get_trigger_memory: Returns the subject’s psychological trigger profile, including active and resolved topics.
• get_session_history: Returns timestamped state records and identifies the trend as rising, falling, or stable.

Nefesh — Real-Time Human State Awareness for AI MCP server is a fit for agents that need to adjust conversation length, tone, or escalation based on incoming human-state data. The provided material does not describe local sensor collection; clients must supply the signal data sent to ingest.