Build somatic-aware AI agents.
Cathexis exposes a Model Context Protocol (MCP) server that streams structured somatic data from the user's device. Your AI agent can access real-time body state, pattern history, and clinical intelligence — with the user's explicit consent and a permission model that enforces clinical safety boundaries.
Everything stays on the user's device. The MCP server runs locally. Connection via Bonjour discovery on local WiFi.
What the MCP server provides
11 tools for querying somatic state, check-in history, pattern analysis, intervention data, prevention signals, timeline summaries, and learning trajectories. 4 resources providing assessment status, ontology definitions, program state, and visualization architecture. SSE streaming for real-time pattern history and prevention signals. JSON schemas for all tool inputs.
Claude Desktop connects natively. Any MCP-compatible client can discover and connect to the Cathexis server running on the user's device.
Built on three unbreakable principles
Local-first
Runs on 127.0.0.1. No cloud relay. No third-party services. Your agent talks directly to the user’s device.
User-controlled
Every connection is visible and manageable inside Cathexis. Users grant, limit, or revoke access with one tap. Revocation is instant and total.
Standards-based
JSON-RPC 2.0 over HTTP/SSE. Zero-config discovery via Bonjour/mDNS. Works natively with Claude, Grok, OpenAI, and any custom MCP client.
Four-level permission model
All data stays on the user's device. The MCP server enforces permission levels that determine what an external agent can access:
L1 — General Wellness: Aggregated engagement metadata — check-in frequency, program progression, feature usage. No somatic content. Sufficient for scheduling, reminders, and general encouragement.
L2 — Somatic Awareness: Current somatic data — zone, quality, intensity — and temporal patterns. Sufficient for contextual responses that reference the user's body state.
L3 — Pattern Intelligence: Pattern detection outputs, territory assignments, cross-domain correlations, prevention signals. Sufficient for clinical-level reasoning about the user's predictive processing patterns. Requires explicit user consent with 30-day expiration.
L4 — Timeline Arc: Full longitudinal arc — 90-day trajectory summaries, learning level progression, prediction accuracy trends. Enables sophisticated therapeutic reasoning about change over time.
Permission changes take effect immediately. The user can revoke access at any time from the app.
Tools
Eleven observational tools across three permission levels. All responses use strictly neutral, body-based language — no diagnoses, no clinical claims. Just the raw signals the nervous system is generating.
-
Current somatic activation, active territory, last check-in summary
-
Learning level, intervention response data
-
Recommended intervention based on current state
-
User's protective factor profile
-
Recent check-ins with zone/quality/intensity data
-
Contextual data, ecological pressure, life transitions
-
Prediction tracking results
-
Pattern detection results
-
Active prevention escalation state
-
90-day rolling window, territory evolution, key transitions
-
Bayesian advancement history, mastery confidence
Cathexis Somatic Ontology (CSO)
A formal vocabulary for representing body-based experience in computational systems. 29 anatomical zones, 20 sensation qualities, 7 primary affect systems, 19 brain regions, 8 functional networks, 5 ecological context layers, and 16 territories of human experience. Organized within a Markov blanket framework (internal, external, sensory, active states) derived from Friston's free energy principle.
Available in OWL, Turtle, and JSON-LD. The JSON-LD context document enables any MCP-compatible agent to produce semantically tagged somatic observations that interoperate with Cathexis data. Cross-mapped to ICD-11, SNOMED CT, and HL7 FHIR.
Connect to Cathexis
1. Install Cathexis on an iOS device
2. Open Settings → AI Integration → Enable MCP Server
3. Your MCP client discovers the server via Bonjour on local WiFi
4. The user approves the connection and sets permission level
5. Query somatic data using the published tool schemas
The app provides connection details and a step-by-step guide under Settings → AI Integration → How to Connect.
Open-source timeline
Current (v1.1): Published on Zenodo, BioPortal, and GitHub. Machine-readable formats available (OWL, Turtle, JSON-LD). Open-source vocabulary specification at github.com/justindegarbo/mcp-somatic.
Post-validation study: Core ontology files open-sourced under CC BY-NC-SA 4.0. Developer SDK published. Community contribution guidelines established.
What is never shared
The following data is architecturally excluded at every permission level:
Substance use data
Medication names, doses, or schedules
Journal entries and free-text reflections
Raw body map coordinates
High-urgency prevention signals
Manic episode trajectory data
Privacy
Cathexis is a wellness app, not a medical device. All pattern data is observational — it describes what the user's body has been showing, not what is clinically true about them. No data is transmitted to Cathexis servers via MCP. The server runs on the user's device. Your agent talks directly to the phone.
Questions? support@cathexis.health