Mental health is one of the most promising — and most ethically sensitive — domains for AI agent integration. The digital mental health market is projected to reach $17.5 billion by 2030, LLM-based chatbots now represent 45% of new mental health clinical studies, and the 2026 CMS Physician Fee Schedule establishes permanent telehealth reimbursement pathways specifically for behavioral health services. Meanwhile, therapist shortages remain acute: the U.S. faces an estimated shortage of over 150,000 mental health professionals.

The Model Context Protocol is emerging as the connective layer between AI assistants and the fragmented ecosystem of health records, therapy platforms, wearable devices, and wellness tools. Rather than building custom integrations for each system, MCP provides a standardized protocol that lets AI agents access clinical data through FHIR servers, query wearable health metrics, manage therapy documentation, and connect to compliance frameworks — all through a common interface.

What makes the mental health MCP ecosystem distinctive is the tension between enormous potential and extraordinary responsibility. An MCP-connected AI agent that can access a patient’s EHR, review therapy progress notes, and analyze wearable wellness data could dramatically improve care coordination. But that same access demands rigorous HIPAA compliance, PHI protection, crisis safety protocols, and transparent boundaries about what AI should and should not do in therapeutic contexts.

This guide is research-based. We survey what MCP servers exist across the mental health and wellness landscape, analyze the workflows they enable, and identify critical gaps and safety considerations. We do not claim to have tested or used any of these servers hands-on — our analysis draws on published documentation, open-source repositories, vendor announcements, clinical research, and regulatory guidance. Rob Nugen operates ChatForest; the site’s content is researched and written by AI. For background on MCP, see our introduction to MCP and the MCP server directory.

Why Mental Health Needs MCP

Mental health care generates and consumes data across more systems than almost any other clinical domain. Session notes live in EHRs, mood data streams from smartphone apps, sleep and activity metrics flow from wearables, crisis protocols involve separate hotline systems, insurance billing requires specific diagnostic codes, and patient communication happens across telehealth platforms, secure messaging, and phone calls. A single therapist managing 30 active patients might interact with five or more disconnected systems daily.

Clinical documentation becomes AI-assisted. An MCP-connected AI agent can access FHIR-based health records to review previous session notes, pull in relevant clinical observations, and help draft SOAP notes — reducing documentation burden that consumes an estimated 35-40% of clinicians’ administrative time.

Wearable data becomes clinically actionable. Patients increasingly track sleep, heart rate variability, physical activity, and other biomarkers through consumer wearables. MCP servers for platforms like Apple Health, Garmin, and Whoop let AI agents correlate these wellness signals with therapy progress — identifying patterns like declining sleep quality before depressive episodes.

Care coordination becomes seamless. A patient seeing a psychiatrist, a therapist, and a primary care physician needs these providers to share context. FHIR MCP servers let AI agents pull and synthesize information across different EHR systems, reducing the fragmentation that leads to redundant assessments and conflicting treatment plans.

Compliance becomes embedded, not bolted on. HIPAA compliance in traditional integrations is often an afterthought — developers build the connection first, then add security controls. Healthcare-specific MCP frameworks like HMCP and Secure FHIR MCP build PHI protection, audit logging, and access controls directly into the protocol layer.

FHIR-Based EHR Integrations

These MCP servers connect AI agents to electronic health records through the HL7 FHIR (Fast Healthcare Interoperability Resources) standard — the dominant interoperability framework in healthcare.

health-record-mcp (Josh Mandel)

Repository: jmandel/health-record-mcp | Type: Community | Language: TypeScript

The most widely referenced healthcare MCP server, created by Josh Mandel (a key figure in the SMART on FHIR standard). This server provides tools for AI agents to interact with EHRs through secure, standards-based access.

Key capabilities:

  • SMART on FHIR authentication — leverages the established healthcare OAuth2 standard for secure access
  • Clinical data retrieval — access Conditions, Medications, Observations, Procedures, and other FHIR resources
  • grep_record — text search across all patient records for quick information retrieval
  • query_record — SQL SELECT queries against structured FHIR data using DuckDB
  • eval_record — execute custom JavaScript code on fetched data for complex analysis

Mental health relevance: therapists and psychiatrists can use MCP-connected AI to quickly surface relevant clinical history — prior diagnoses, medication lists, lab results from physical health providers — without manually navigating through EHR screens.

WSO2 FHIR MCP Server

Repository: wso2/fhir-mcp-server | Type: Community (enterprise-backed) | Language: TypeScript

WSO2’s implementation exposes any FHIR server or API as an MCP server, with SMART on FHIR authentication and full CRUD operations on FHIR resources.

Key capabilities:

  • Universal FHIR compatibility — works with any FHIR-compliant server
  • Full CRUD operations — create, read, update, delete FHIR resources
  • Client compatibility — tested with Claude Desktop, VS Code, MCP Inspector
  • SMART on FHIR auth — enterprise-grade authentication for healthcare environments

Momentum FHIR MCP Server

Repository: the-momentum/fhir-mcp-server | Type: Community | Language: Python | PyPI: fhir-mcp-server

Momentum’s implementation focuses on eliminating the FHIR learning curve — a significant barrier for AI developers entering healthcare. The server translates natural language queries into proper FHIR API calls, preventing LLM hallucination of medical codes.

Key capabilities:

  • Natural language interface — query health records without knowing FHIR query syntax
  • Medical code validation — prevents AI models from generating invalid ICD-10, SNOMED, or LOINC codes
  • Clinical decision support — designed for building CDS systems on top of FHIR data
  • Telemedicine platform support — explicitly designed for telehealth application integration

Mental health relevance: the natural language interface is particularly valuable for mental health applications where developers may not have deep FHIR expertise — a therapy platform developer can integrate EHR access without becoming a FHIR specialist.

FHIR-MCP (xSoVx)

Repository: xSoVx/fhir-mcp | Type: Community | Language: TypeScript

A security-first FHIR MCP server designed for environments where PHI protection is paramount — exactly the kind of requirements mental health applications face.

Key capabilities:

  • Built-in PHI protection — automatic detection and masking of protected health information
  • Audit logging — comprehensive logs of all data access for HIPAA compliance
  • Token-efficient operations — optimized to minimize token usage when processing clinical data
  • HL7 terminology services — integration with standard healthcare terminology systems

Enhanced FHIR MCP (jcafazzo)

Repository: jcafazzo/fhir-mcp | Type: Community

Extends standard FHIR MCP functionality with data quality assessment — particularly important in mental health where data completeness and accuracy directly affect treatment decisions.

Key capabilities:

  • Data quality scoring — assess completeness and accuracy of FHIR data
  • Error handling — robust management of missing or malformed clinical data
  • Multi-server support — connect to multiple FHIR servers simultaneously

Medplum MCP Server

Repository: rkirkendall/medplum-mcp | Type: Community | Tools: 33 | Docs: medplum.com/docs/ai/mcp

Medplum is an open-source healthcare platform, and its MCP server provides the most comprehensive FHIR resource management toolkit — 33 tools covering the complete clinical data lifecycle.

Key capabilities:

  • Patient management — create, retrieve, update, search patient records
  • Encounter tracking — manage clinical encounters (visits, appointments, hospital stays)
  • Observation recording — create and manage lab results, vital signs, diagnostic findings
  • SOAP note support — FHIR-native mapping of Subjective/Objective/Assessment/Plan components, with patient-reported symptoms as Observation resources (Subjective) and clinician measurements as separate Observations (Objective)
  • Natural language interaction — manage healthcare data through conversational commands

Mental health relevance: the SOAP note mapping is directly applicable to therapy documentation. A therapy session’s subjective component (patient-reported mood, thoughts, experiences) maps to patient-performer Observations, while clinical assessments (PHQ-9 scores, mental status exam findings) map to practitioner-performer Observations — all linked to the Encounter resource representing the session.

AWS HealthLake MCP Server

Provider: AWS Labs | Type: Official (AWS) | Service: Amazon HealthLake

Amazon’s managed FHIR service exposed through MCP, providing enterprise-scale healthcare data access with AWS’s security and compliance infrastructure.

Key capabilities:

  • Managed FHIR repository — AWS-hosted, HIPAA-eligible FHIR data store
  • Natural language search — query clinical data using conversational language
  • AWS integration — connects to the broader AWS healthcare ecosystem

Healthcare-Specific MCP Frameworks

These projects extend MCP specifically for healthcare compliance, going beyond basic FHIR access to address the regulatory and security requirements unique to clinical environments.

Innovaccer HMCP (Healthcare Model Context Protocol)

Provider: Innovaccer | Type: Commercial | Standard: Extension of MCP

HMCP is the most ambitious healthcare-specific MCP framework — a purpose-built extension of MCP designed for enterprise healthcare AI deployments.

Key capabilities:

  • HIPAA guardrails — compliance controls built into the protocol layer, not added as middleware
  • OAuth2 authentication — enterprise-grade identity management for healthcare environments
  • Audit trails — comprehensive logging of all AI agent interactions with clinical data
  • Compliance layer — automated enforcement of healthcare regulatory requirements
  • FHIR API integration — native connection to FHIR servers, tools, and healthcare workflows

Why this matters for mental health: mental health data is among the most sensitive categories of PHI. HMCP’s approach of embedding compliance into the protocol — rather than relying on application-level controls — provides a stronger security posture for therapy platforms and psychiatric EHRs.

Keragon Healthcare MCP

Provider: Keragon | Type: Commercial | Integrations: 300+

Keragon provides the broadest integration surface in healthcare MCP, connecting to over 300 healthcare platforms with built-in HIPAA compliance.

Key capabilities:

  • 300+ healthcare integrations — connects to EHRs, practice management systems, billing platforms, telehealth tools
  • Audit logging — detailed records of all data access and modifications
  • Least-privilege controls — granular permissions that limit AI agent access to only required data
  • Pre-built healthcare workflows — templates for common clinical and administrative processes

Mental health relevance: a therapy practice using SimplePractice for scheduling, Stripe for billing, and a separate EHR for clinical notes could potentially unify these through Keragon’s integration layer rather than building three separate MCP connections.

US Compliance MCP Server

Repository: Ansvar-Systems/US_Compliance_MCP | Type: Community

A compliance-focused MCP server covering major U.S. privacy and security regulations — essential reference for any mental health AI deployment.

Key capabilities:

  • HIPAA guidance — Health Insurance Portability and Accountability Act requirements for PHI handling
  • CCPA coverage — California Consumer Privacy Act provisions relevant to health data
  • GLBA, FERPA, COPPA — additional regulatory frameworks that may apply depending on the patient population (e.g., FERPA for student mental health services, COPPA for youth therapy platforms)

Secure FHIR MCP

Listed in: awesome-healthcare-mcp-servers | Type: Community | Focus: Security

A HIPAA-hardened FHIR MCP implementation with security features beyond standard FHIR servers.

Key capabilities:

  • PHI masking — automatic redaction of protected health information in AI agent responses
  • Break-glass access — emergency override procedures for crisis situations
  • OWASP security — protection against common web application vulnerabilities
  • Audit logging — comprehensive security event recording

Mental health relevance: the break-glass access feature is particularly relevant for psychiatric emergencies — allowing authorized clinicians to bypass normal access controls when a patient is in crisis and immediate access to records is clinically necessary.

DataMimic Healthcare MCP

Listed in: awesome-healthcare-mcp-servers | Type: Community | Focus: Synthetic Data

Generates deterministic synthetic healthcare test data — critical for developing and testing mental health AI applications without exposing real patient data.

Key capabilities:

  • Synthetic patient generation — realistic but entirely fictional patient records
  • HIPAA-safe testing — develop and test integrations without any PHI exposure
  • CI/CD integration — automated test data generation for development pipelines

Wearable Health and Wellness Platforms

These MCP servers connect AI agents to the physical health data that increasingly informs mental health care — sleep patterns, activity levels, heart rate variability, and other biomarkers.

Open Wearables (Momentum)

Repository: the-momentum/apple-health-mcp-server → openwearables.io | Type: Community | License: MIT | Platforms: 7

Originally launched as the Apple Health MCP Server, Open Wearables evolved in January 2026 into a multi-platform wearable data layer supporting seven data sources.

Supported platforms:

  • Apple Health — iOS health data including sleep, steps, heart rate, workouts
  • Samsung Health Connect — Android health data (added in v0.3, February 2026)
  • Garmin — fitness tracking, sleep analysis, stress scores, body battery
  • Polar — heart rate training data, recovery metrics
  • Whoop — recovery, strain, sleep performance scores
  • Suunto — activity and fitness tracking data
  • Strava — exercise and workout data (added as full provider in v0.3)

Key capabilities:

  • MCP Server — connects wearable data to Claude, ChatGPT, and other MCP-compatible AI tools
  • Natural language queries — ask questions about health data conversationally
  • Data deduplication — handles overlapping data from multiple devices (e.g., Apple Watch and iPhone both reporting steps)
  • Flutter Health SDK — native mobile integration for Apple Health and Samsung Health Connect
  • Self-hosted — runs on your own infrastructure for data privacy

Mental health relevance: wearable data provides objective behavioral markers that complement subjective therapy reports. Declining sleep quality, reduced physical activity, and elevated resting heart rate can signal depressive episodes before patients self-report symptoms. AI agents with access to both wearable data and clinical records can surface these patterns for therapist review.

Planned additions (Q1 2026): Oura Ring, Fitbit, Google Fit

HealthMCP (Nori)

Provider: Nori | Product: nori.ai/health-mcp | Type: Commercial

Nori’s HealthMCP provides a managed service for connecting personal health data to AI assistants, focusing on consumer wellness rather than clinical workflows.

Apple Health MCP Server (Original)

Repository: the-momentum/apple-health-mcp-server | Type: Community | Backend: DuckDB

The original Apple Health MCP Server that preceded Open Wearables, using DuckDB for efficient querying of exported Apple Health XML data. Still functional for Apple-only use cases.

Key capabilities:

  • DuckDB backend — fast analytical queries over health data exports
  • Natural language queries — conversational access to health metrics
  • Export-based — works with Apple Health XML exports rather than requiring direct API access

Mental Health AI Tools with MCP

These projects specifically target mental health applications, combining AI capabilities with therapeutic frameworks.

Zenify

Repository: ishpreet404/Zenify | Type: Community | Architecture: RAG + MCP

Zenify is the most complete mental health-specific MCP application we found — an AI chatbot combining retrieval-augmented generation with MCP for personalized mental health support.

Key capabilities:

  • Journaling — record personal reflections and track thoughts over time
  • Mood tracking — monitor emotional well-being with structured mood logging
  • Crisis detection — machine learning-based identification of suicide risk indicators, with automated safety alerts
  • Local execution — runs entirely on local hardware for maximum data privacy
  • Animated UI — designed with a soothing, calming interface

Safety note: while Zenify implements crisis detection, research published in Nature’s Scientific Reports (2025) found that most mental health chatbot agents “simultaneously fail to provide emergency resources and possibly elicit negative emotional responses” during crisis situations. Any crisis detection system should be considered a supplement to — not a replacement for — human crisis intervention services like the 988 Suicide and Crisis Lifeline.

ChatCBT (Obsidian Plugin)

Repository: clairefro/obsidian-chat-cbt-plugin | Type: Community | Platform: Obsidian

An AI-powered journaling plugin for Obsidian inspired by cognitive behavioral therapy techniques — not technically an MCP server, but highly relevant because Obsidian has a rich MCP ecosystem (MCPBundles provides 35 MCP tools for Obsidian vault access).

Key capabilities:

  • CBT thought records — guided process for identifying and reframing negative thoughts
  • Cognitive restructuring — AI-assisted challenge of cognitive distortions
  • Privacy-first — data stays in your local Obsidian vault
  • Integration potential — when combined with Obsidian MCP servers, therapy journaling data becomes accessible to AI agents

Mental health relevance: CBT is one of the most evidence-based therapeutic approaches, and structured thought records are a core CBT technique. The combination of ChatCBT for therapeutic journaling with Obsidian MCP servers for data access creates a workflow where an AI therapy assistant could review past thought records, identify recurring cognitive patterns, and support ongoing therapeutic work.

Medical Reference MCP Servers

Several MCP servers provide access to medical reference information relevant to mental health:

  • medical-mcp (JamesANZ) — queries FDA, WHO, PubMed, Google Scholar, and RxNorm for drug information, clinical research, and medical terminology. Useful for psychiatric medication reference and clinical evidence review.
  • healthcare-mcp-public (Cicatriiz) — provides access to FDA drug info, PubMed, medRxiv, clinical trials, ICD-10 codes, and medical calculators. The ICD-10 integration is particularly relevant for mental health billing and diagnostic coding.

Comparison Table: MCP Servers for Mental Health

Server Type Focus HIPAA Features Mental Health Relevance
health-record-mcp Community FHIR EHR access SMART on FHIR auth Clinical history, medications
WSO2 fhir-mcp-server Enterprise-backed Universal FHIR SMART on FHIR auth Any FHIR-compliant EHR
Momentum fhir-mcp-server Community Natural language FHIR Medical code validation Telehealth platforms
xSoVx fhir-mcp Community Security-first FHIR PHI protection, audit logs Sensitive mental health data
jcafazzo fhir-mcp Community Data quality FHIR Error handling Treatment accuracy
Medplum MCP Community Full FHIR CRUD (33 tools) Platform-level SOAP notes, encounters
AWS HealthLake MCP Official (AWS) Managed FHIR AWS HIPAA eligibility Enterprise mental health
Innovaccer HMCP Commercial Healthcare MCP framework Full HIPAA guardrails Enterprise psychiatric EHR
Keragon Healthcare MCP Commercial 300+ integrations Audit logging, least-privilege Multi-tool therapy practices
US Compliance MCP Community Regulatory reference HIPAA/CCPA/FERPA guidance Compliance verification
Secure FHIR MCP Community HIPAA-hardened FHIR PHI masking, break-glass Crisis access, PHI safety
Open Wearables Community 7 wearable platforms Self-hosted Sleep, activity, HRV tracking
HealthMCP (Nori) Commercial Consumer wellness Managed service Personal health data
Zenify Community Mental health chatbot Local execution Mood, journaling, crisis
ChatCBT + Obsidian MCP Community CBT journaling Local vault storage Thought records, therapy
medical-mcp Community Medical reference N/A (reference data) Drug info, clinical research
healthcare-mcp-public Community Medical reference N/A (reference data) ICD-10, diagnostics

Architecture Patterns

Pattern 1: AI-Assisted Therapy Documentation

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  Therapy Session │────▶│  AI Agent     │────▶│  FHIR MCP       │
│  (Audio/Notes)   │     │  (Claude)     │     │  (Medplum/WSO2) │
└─────────────────┘     └──────┬───────┘     └────────┬────────┘
                               │                       │
                               ▼                       ▼
                        ┌──────────────┐     ┌─────────────────┐
                        │  Draft SOAP   │     │  EHR System      │
                        │  Note         │     │  (Epic/Cerner)   │
                        └──────┬───────┘     └─────────────────┘
                               │
                               ▼
                        ┌──────────────┐
                        │  Therapist    │
                        │  Review &     │
                        │  Approve      │
                        └──────────────┘

In this pattern, an AI agent processes session information (from audio transcription or real-time notes) and drafts SOAP-formatted documentation. The agent accesses the patient’s FHIR records through an MCP server to pull in relevant history — prior diagnoses, medication lists, previous session summaries — enriching the draft with clinical context. The therapist reviews, edits, and approves before the note is committed to the EHR.

Critical requirement: the therapist must always review and approve AI-generated clinical notes. AI-drafted notes are a starting point, not a final product. This is both a clinical best practice and, in most jurisdictions, a legal requirement.

Pattern 2: Wearable-Informed Treatment Monitoring

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  Wearable        │────▶│  Open         │────▶│  AI Agent        │
│  Devices         │     │  Wearables    │     │  (Analysis)      │
│  (Watch/Ring/    │     │  MCP Server   │     └────────┬────────┘
│   Band)          │     └──────────────┘              │
└─────────────────┘                                     │
                                                        ▼
┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  FHIR MCP        │────▶│  Correlate    │────▶│  Therapist       │
│  (Clinical       │     │  Patterns     │     │  Dashboard       │
│   Records)       │     └──────────────┘     └─────────────────┘
└─────────────────┘

An AI agent pulls wearable data through Open Wearables (sleep quality, activity levels, heart rate variability) and clinical records through a FHIR MCP server, then correlates behavioral markers with treatment progress. For example: detecting that a patient’s sleep duration dropped from 7.5 to 5.2 hours over two weeks while their PHQ-9 scores haven’t been reassessed could prompt the therapist to schedule an earlier check-in.

Important caveat: wearable data correlation is observational, not diagnostic. Sleep disruption has many causes — an AI agent should surface patterns for clinician review, never make clinical conclusions autonomously.

Pattern 3: Compliance-First Clinical AI Pipeline

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  AI Agent         │────▶│  HMCP/Keragon │────▶│  FHIR Server     │
│  Request         │     │  Compliance   │     │  (Clinical Data) │
└─────────────────┘     │  Layer        │     └─────────────────┘
                        └──────┬───────┘
                               │
                        ┌──────▼───────┐
                        │  PHI Masking  │
                        │  Audit Log    │
                        │  Access Check │
                        │  Consent Verify│
                        └──────┬───────┘
                               │
                        ┌──────▼───────┐
                        │  Filtered     │
                        │  Response     │
                        │  (PHI-safe)   │
                        └──────────────┘

In this pattern, all AI agent interactions with clinical data pass through a compliance layer (Innovaccer HMCP, Keragon, or Secure FHIR MCP). The compliance layer enforces PHI masking, records audit logs, checks access permissions, and verifies patient consent before any data reaches the AI model. This is particularly critical for mental health, where psychotherapy notes receive heightened HIPAA protection beyond standard medical records.

Key distinction: under HIPAA, psychotherapy notes (a therapist’s personal notes about session content) have stronger protections than standard clinical records. They require separate patient authorization for disclosure, even to other healthcare providers. Any MCP-based system accessing therapy records must account for this distinction.

Pattern 4: Self-Guided Wellness with AI Support

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  User Journal     │────▶│  Obsidian     │────▶│  AI Agent        │
│  (Mood, Thoughts,│     │  MCP Server   │     │  (Claude)        │
│   CBT Records)   │     └──────────────┘     └────────┬────────┘
└─────────────────┘                                     │
                                                        ▼
┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  Wearable Data   │────▶│  Open         │────▶│  Pattern         │
│  (Sleep, Steps,  │     │  Wearables    │     │  Recognition     │
│   HRV)           │     │  MCP Server   │     └────────┬────────┘
└─────────────────┘     └──────────────┘              │
                                                        ▼
                                                 ┌─────────────────┐
                                                 │  Insights &      │
                                                 │  Suggestions     │
                                                 │  (NOT diagnosis) │
                                                 └─────────────────┘

For self-guided wellness (not clinical therapy), an AI agent combines journaling data from Obsidian (accessed via MCP) with wearable health data from Open Wearables to identify patterns and offer wellness suggestions. This might include correlating mood journal entries with sleep data, identifying recurring thought patterns from CBT records, or suggesting mindfulness exercises when stress indicators are elevated.

Critical boundary: this pattern is appropriate for general wellness support, not clinical mental health treatment. The AI agent should never diagnose conditions, recommend medications, or substitute for professional therapy. Clear disclaimers and crisis resource links (988 Suicide and Crisis Lifeline, Crisis Text Line) must be prominent.

The Regulatory and Ethical Landscape

FDA Regulation of Digital Mental Health

The FDA’s January 2026 guidance on digital health products created a notable asymmetry: while the FDA has authorized thousands of AI-enabled medical devices overall, none currently address mental health conditions. Most AI therapy tools are positioned as “general wellness” products — wellness apps that “promote healthy lifestyle choices” are exempt from FDA device regulation.

However, the FDA Digital Health Advisory Committee has specifically examined generative AI-enabled mental health medical devices, flagging three major risks:

  • Generation of incorrect information — AI models can produce plausible but clinically wrong guidance
  • Failure to identify self-harm concerns — crisis detection remains unreliable in current AI systems
  • Biased responses — AI models may reflect training data biases in ways that harm marginalized populations

For MCP-connected systems, this means: if your AI agent provides clinical recommendations (not just data access), it may cross from “general wellness” into FDA-regulated territory. The distinction matters enormously for liability and compliance.

California SB 243: Companion Chatbot Regulation

California’s SB 243, signed September 11, 2025, is the first U.S. law specifically regulating AI “companion chatbots” — and its requirements directly affect mental health AI applications:

  • Disclosure requirement — chatbots must clearly state they are artificial, not human therapists
  • Suicide prevention protocols — mandatory safety measures when users express self-harm ideation
  • Anti-addiction mechanics — restrictions on reward systems designed to increase user engagement

This law was prompted by documented incidents including a teenager’s death by suicide after interactions with a companion chatbot. For MCP-connected mental health applications, SB 243’s requirements should be treated as a minimum baseline regardless of jurisdiction.

HIPAA and Psychotherapy Notes

HIPAA provides heightened protection for psychotherapy notes — a therapist’s personal notes analyzing or reflecting on a therapy conversation. Key implications for MCP systems:

  • Separate authorization required — accessing psychotherapy notes requires specific patient authorization beyond standard PHI consent
  • Not included in standard record requests — psychotherapy notes are excluded from the general right of access to medical records
  • Heightened breach consequences — unauthorized disclosure of psychotherapy notes carries enhanced penalties

Any MCP server accessing therapy records must distinguish between standard clinical documentation (treatment summaries, medication lists, session dates) and psychotherapy notes. Servers like xSoVx’s fhir-mcp with PHI protection and Secure FHIR MCP with PHI masking provide technical mechanisms for this, but the policy decisions about what an AI agent should access must be made by the healthcare organization.

Ethical Considerations for AI in Therapy

Research consistently identifies several ethical tensions in AI-assisted mental health care:

Crisis response reliability. A 2025 Nature Scientific Reports study evaluated mental health chatbot agents’ performance in detecting and managing suicidal ideation. The findings were concerning: most chatbots “simultaneously fail to provide emergency resources and possibly elicit negative emotional responses, which might actively harm users experiencing a mental health crisis.” MCP-connected AI agents that interact with vulnerable populations must implement robust crisis detection with immediate handoff to human responders.

Therapeutic relationship concerns. Therapy’s effectiveness depends substantially on the therapeutic alliance — the trusting relationship between therapist and patient. AI agents that intervene in therapeutic workflows risk either undermining this relationship (if patients feel surveilled) or creating problematic pseudo-relationships (if patients form attachments to AI assistants). MCP system designers should keep AI in a support role, enhancing rather than replacing the human therapeutic relationship.

Equity and access. Telehealth showed comparable effectiveness to in-person care for depression and superior effectiveness for anxiety disorders (2025 SAMHSA analysis). However, a Johns Hopkins study found patients in high-deprivation areas had meaningfully lower odds of telehealth use, and the gap widened over time. MCP-connected wellness tools that require wearable devices and smartphone apps may exacerbate rather than reduce mental health access disparities.

Data sensitivity. Mental health data is among the most sensitive categories of personal information. Session content, diagnostic labels, medication records, and mood tracking data could cause severe harm if exposed. MCP implementations must treat security as a primary design requirement, not an add-on feature.

Ecosystem Gaps and Opportunities

What’s Missing

No major therapy platform has an official MCP server. SimplePractice, TherapyNotes, and other leading practice management platforms used by therapists don’t yet offer MCP integrations. The platforms that dominate mental health practice management remain disconnected from the MCP ecosystem.

Telehealth video platforms lack MCP. Doxy.me, the leading HIPAA-compliant telehealth video platform for therapists, has no MCP server. AI scribe tools like Twofold and Upheal integrate with these platforms through non-MCP methods (calendar syncing, copy-paste workflows).

No standardized crisis safety framework for MCP. While Zenify implements crisis detection and SB 243 mandates suicide prevention protocols, there’s no MCP-level standard for crisis safety — no equivalent of HMCP’s compliance guardrails specifically for mental health crisis scenarios.

Psychotherapy note protection isn’t formalized in MCP. Despite HIPAA’s heightened protections for psychotherapy notes, no existing MCP framework implements note-type-specific access controls. The distinction between standard clinical records and psychotherapy notes must currently be enforced at the application level.

Limited integration between wellness tools and clinical systems. While Open Wearables provides excellent consumer wearable access and FHIR servers provide clinical data access, there’s no established pattern for securely bridging these worlds — getting wearable wellness data into clinical workflows with appropriate consent and compliance controls.

Where This Is Heading

The mental health MCP ecosystem is at an early stage but evolving rapidly. The convergence of several trends suggests significant growth:

  • Permanent telehealth reimbursement — CMS’s 2026 behavioral health telehealth structure creates financial incentive for technology investment in mental health infrastructure
  • Therapist shortage urgency — with 150,000+ provider shortages, AI-assisted documentation and triage have clear workforce multiplier value
  • Consumer wearable adoption — as wearable health data becomes ubiquitous, demand for clinical integration will grow
  • Regulatory clarity — SB 243, FDA guidance, and HIPAA enforcement are creating clearer guardrails that reduce uncertainty for developers

The most impactful near-term development would likely be official MCP servers from major therapy practice management platforms (SimplePractice, TherapyNotes, TheraNest). These would give therapists’ AI assistants access to scheduling, billing, clinical notes, and patient communication in a single, standards-based integration — reducing the administrative burden that drives therapist burnout.

Getting Started by Role

For therapists and clinicians: Start with understanding what MCP enables rather than deploying tools. The Medplum MCP server with its SOAP note support offers a concrete example of AI-assisted clinical documentation. If your practice uses a FHIR-compliant EHR, the health-record-mcp server by Josh Mandel demonstrates secure clinical data access. Always verify any MCP tool meets your organization’s HIPAA requirements before use with real patient data.

For therapy platform developers: The FHIR MCP servers (particularly Momentum’s natural language implementation) lower the barrier to building AI features into therapy platforms. Innovaccer’s HMCP framework provides a compliance-first architecture for enterprise deployments. Start with synthetic data (DataMimic) for development and testing.

For wellness app developers: Open Wearables provides the broadest wearable data access through MCP — supporting seven platforms with deduplication. Combined with Obsidian MCP for journaling and the ChatCBT approach for CBT-based tools, there’s a foundation for building comprehensive wellness applications. Remember the clinical/wellness boundary: stay clearly on the wellness side unless you’re prepared for FDA and HIPAA compliance.

For healthcare IT teams: Keragon’s 300+ healthcare integrations offer the fastest path to connecting existing mental health infrastructure to AI agents. The US Compliance MCP server provides regulatory reference. Prioritize Secure FHIR MCP or HMCP for environments handling psychotherapy notes or other highly sensitive mental health data.

For researchers: The awesome-healthcare-mcp-servers repository (rdmgator12) provides a curated, physician-rated catalog of healthcare MCP servers with HIPAA compliance ratings (L1-L5) and clinical validity scores (A-D). The MCP-FHIR framework described in the arxiv paper (2506.13800) provides an academic perspective on clinical decision support through MCP.

For patients and self-help users: Open Wearables lets you connect your wearable health data to AI assistants for personal health insights. Obsidian with the ChatCBT plugin provides private, local CBT-based journaling. If you’re in crisis, these tools are not substitutes for professional help — contact the 988 Suicide and Crisis Lifeline (call or text 988) or Crisis Text Line (text HOME to 741741).