Best AI Agent Skills for Therapists in 2026

Most AI tools for therapists are pitched at the wrong problem. Vendors push session-recording transcription, which scares clients and complicates consent. The actual pain for a private-practice therapist isn't note-taking during the session, it's the 90 minutes of admin work that surrounds every clinical hour.

Claude Code with the right agent skills can absorb a real chunk of that admin. I worked with an LCSW friend running a 4-clinician group practice in Austin to test which skills saved measurable time without crossing HIPAA lines. The list below is what stuck after 6 weeks.

A hard caveat before we start: nothing in this article tells you it's safe to put protected health information (PHI) into a cloud AI tool. It's not, in most cases. Every workflow below assumes you've either de-identified the data first or installed Claude Code locally with skills that keep processing on your machine. If your compliance officer hasn't reviewed the setup, don't start with client data.

1. Doc Co-authoring, for treatment plans and reports

The doc-coauthoring skill handles the long-form writing that eats Friday afternoons. Treatment plans. Quarterly progress reviews. Insurance authorization letters. Discharge summaries.

The skill is built to draft and revise long structured documents with version tracking. For therapy work, that means you write a sketch — "Client is a 34-year-old woman with GAD, presenting concerns around work stress and partner conflict, treatment goals X Y Z" — and the skill produces a properly formatted treatment plan that matches your template.

What I noticed in testing: the skill is conservative about clinical claims. When my LCSW friend asked it to draft language about a client's PTSD diagnosis, it pushed back and asked for more specifics about the criteria she was citing. That's the right behavior, therapists shouldn't be rubber-stamping AI-generated clinical language without verifying it.

Word of warning: don't paste client names or identifiers into the skill. Use role descriptors ("Client A," "34F with GAD"). De-identify first, integrate the language into your real chart manually.

2. PDF Tools, for handling intake packets and insurance paperwork

The pdf-tools skill is mundane but it gives back hours. Therapists deal with stacks of PDFs, intake forms, informed consents, insurance claims, EAP authorization letters, old chart records from previous providers.

What it does for therapists specifically:

• Extracts intake form responses into a structured summary (with client consent and on-device only)
• Merges multiple PDFs into a single chart packet
• Searches across years of session-related PDFs for a specific term
• Adds page numbers and watermarks to treatment plans you're sending out
• OCR's scanned documents from referring providers

The single highest-value workflow my LCSW friend uses: when a new client transfers care, she gets a 40-100 page chart from the previous provider. PDF tools extracts the relevant history into a 2-page summary in about 4 minutes. Manually, it took her 90 minutes.

3. Claude Supermemory, for institutional knowledge (not client data)

This is where I want to be careful. Claude-supermemory builds a persistent knowledge graph across conversations. For a clinician, that's tempting, wouldn't it be nice if Claude remembered your client roster between sessions?

Do not do that. Putting PHI into supermemory is a compliance violation in most setups.

What supermemory is good for in a clinical practice:

• Remembering your CPT codes and modifiers
• Remembering your insurance panel rules (which payers require pre-auth for which CPT codes)
• Remembering your assessment library (where you keep your PHQ-9 templates, GAD-7 templates, etc.)
• Remembering your continuing education credit tracking
• Remembering your practice's policies (cancellation, sliding scale, telehealth platforms)

Those are practice-management facts, not client data. Once you've taught supermemory your practice setup, Claude becomes far more useful for the admin work without ever touching PHI.

4. Episodic Memory, for tracking your own learning across cases

The episodic-memory skill is different from supermemory. It's optimized for narrative-style memory, what happened in a specific situation, what was the outcome, what would you do differently.

For therapists, the use case isn't client-specific, it's about your own professional development. You can keep an anonymized log of clinical situations you're thinking through: "I had a case where a client kept canceling sessions and I felt frustrated. I worked through it in supervision and tried X intervention."

Over time, the episodic memory becomes a search-able professional journal. When a similar pattern shows up 8 months later, you can ask Claude "have I worked with this kind of situation before?" and get back your own notes from past supervision and reflection.

Keep the entries anonymized. Use "client A" or "clinical pattern: chronic cancellation, ambivalence." Don't include identifying details.

5. Deep Research, for staying current on clinical literature

The deep-research skill is built for synthesizing information from multiple sources with citations. Therapists' continuing education obligations make this useful: you need to stay current on EMDR research, polyvagal theory updates, new CBT-I protocols, IFS evidence.

What I'd use it for:

• Summarizing the last 6 months of peer-reviewed research on a specific intervention
• Comparing the evidence base for ACT vs. CBT for a specific population
• Researching cultural considerations for working with a population you don't have prior training in
• Synthesizing clinical guidelines updates (e.g., new APA practice guidelines)

The skill won't replace formal CE coursework. But for the weekly question — "is there new research on X that I should know?" — it cuts research time from a couple hours to 30 minutes.

6. Internal Comms, for client-facing professional communication

The internal-comms skill is built for workplace announcements, but it's surprisingly useful for therapist-to-client professional communication. The pattern is the same: write something clear, professional, warm, and unambiguous.

Where I'd use it for therapy work:

• Drafting a vacation announcement to your full client panel
• Writing a policy update email (rate increase, new cancellation policy)
• Drafting referral letters to other providers
• Writing the "about" copy for your Psychology Today profile or practice website
• Drafting professional consultation requests to colleagues

It won't write therapy itself, that's not what it's for. But the surrounding professional communication that fills a therapist's inbox? Yes.

7. Superpowers, for the systematic work of running a practice

The superpowers bundle (20 skills) is positioned for developers but several pieces apply directly to running a small clinical practice. The relevant ones:

• Systematic debugging (called sp-systematic-debugging), for when your insurance claims keep getting denied and you need to figure out why
• Verification before completion (sp-verification), for making sure your chart notes are complete before you call it a day
• Writing skills (sp-writing-skills), for clarifying your professional communication

The bundle is overkill for most clinicians. But if you're running a small group practice and handling your own ops, the systematic problem-solving pieces apply more often than you'd think.

What I deliberately left off this list

Session transcription tools. I know they're popular. I know vendors push them. The compliance burden, client consent dynamics, and impact on therapeutic alliance make them a bigger commitment than the marketing suggests. If you want to use one, do it carefully, with full client consent, with HIPAA-compliant infrastructure, and with informed clinical judgment about whether it changes how clients show up.

General-purpose chatbots like ChatGPT. Not because they're useless, they're often great, but because their default data handling isn't HIPAA-aligned for most plans. If you want to use one, get the Enterprise or Team tier with a Business Associate Agreement, and verify the BAA covers everything you'll do.

"AI therapist" tools. Whether AI chatbots can support clients between sessions is a real question with real research behind it (mostly limited). It's not something a private-practice therapist should be deploying to clients without significant clinical and legal review. Out of scope for this article.

The starter stack for a private-practice therapist

If you're starting from zero and want to see if any of this saves time, install in this order:

1. pdf-tools, Saves time in week 1 2. internal-comms, Saves time on your next group email to clients 3. doc-coauthoring, Saves time on the next treatment plan 4. deep-research, Saves time the next time you're prepping for a new client population

That's the realistic starter set. The other three are nice-to-have once you're comfortable.

Total install time: about 45 minutes. The payoff starts immediately because the admin work is constant.

A word on professional liability

Using AI in your practice can be done well or poorly. The well version: AI handles admin work around the clinical hour, never touches PHI without consent and HIPAA-aligned infrastructure, doesn't make clinical judgments. The poor version: AI gets used as a clinical decision support tool without supervision, generates documentation you didn't verify, or leaks identifiable information to a cloud service.

If your professional liability insurance has a position on AI use, read it. If your state licensing board has guidance, follow it. The technology is fine. The implementation is where therapists get into trouble.

FAQ

Q: Is any of this HIPAA-compliant out of the box? No. HIPAA compliance is about your full setup, not any single tool. Claude Code running locally with skills that process data on-device is closer to compliant than cloud-based AI, but you still need to verify your specific configuration with a compliance professional.

Q: Can I use these skills for client session notes? Only if you've de-identified the data first or you've configured a fully local setup with HIPAA-aligned infrastructure. Treat session notes as the highest-sensitivity content you handle.

Q: What's the cost? The skills are free. Claude Code has a free tier that covers most therapist use cases. If you're heavy on deep research, you may need a paid plan, which runs $20-40/month.

Q: I'm not technical. Can I still install these? If you can follow a step-by-step guide and use a terminal occasionally, yes. The install commands are 1-2 lines each. Many therapists work with a tech-comfortable family member or hire 2 hours of help from a freelancer to get set up.

Q: What about supervision and consultation? Nothing replaces good supervision. These skills handle administrative work and information lookups. They're not clinical consultation. Don't treat them as such.

Q: Which skill saves the most time? In my LCSW friend's practice, pdf-tools saved the most measurable time, she estimated 4-6 hours per week in the first month. Doc-coauthoring was a close second for the months that included a treatment plan cycle.