intakeq-mcp
Enables Claude to access IntakeQ practice management data including scheduling, intake forms, treatment notes, invoices, and client records with HIPAA-compliant audit logging.
README
@oktopeak/intakeq-mcp
Built by Oktopeak — AI transformation & automation for healthcare practices
Digital transformation for healthcare and legal businesses. We build AI integrations, workflow automation, and custom software your practice owns outright — including this connector. → Book a 30-min call
IntakeQ/PracticeQ MCP connector for Claude. Gives Claude access to your IntakeQ scheduling, intake forms, treatment notes, invoices, and client records through the Model Context Protocol.
Every PHI read and write is logged locally in compliance with HIPAA §164.312(b).
⚠️ Before you connect this to real patient data
Two things must be in place before PHI flows through this connector:
-
IntakeQ BAA — IntakeQ/PracticeQ includes a Business Associate Agreement. Sign it in your IntakeQ account (Settings → HIPAA) before going live. This is one of IntakeQ's strengths: the BAA is self-serve and included in all plans.
-
Anthropic enterprise tier with ZDR — With hosted Claude + MCP, PHI goes to Anthropic for inference. The defensible claim is training-disabled + Zero Data Retention under enterprise + BAA — not data locality. "Access" (the model processed it) is the exposure, not just retention. Sign up for Anthropic enterprise and request a BAA before using this with real patients.
What this connector does:
- Brokers authenticated calls to the IntakeQ REST API
- Logs every PHI read and write to
~/.intakeq-mcp/audit.log(HIPAA §164.312(b)) - Stores your API key encrypted (AES-256-GCM) locally — no PHI content stored (audit entries contain only access-trail identifiers)
Installation
npm install -g @oktopeak/intakeq-mcp
Claude Desktop configuration
{
"mcpServers": {
"intakeq": {
"command": "intakeq-mcp"
}
}
}
Authentication
Generate your API key in IntakeQ: Settings → Integrations → Developer API
Then tell Claude:
"Set my IntakeQ API key to [your-key]"
Or use the MCP tool directly: set_api_key
Tools
| Tool | Description | R/W |
|---|---|---|
set_api_key |
Store IntakeQ API key | W |
auth_status |
Verify API connection | R |
clear_api_key |
Remove stored key | W |
list_clients |
Search/list clients | R |
get_client |
Single client profile | R |
list_practitioners |
All practitioners in the account | R |
list_services |
Services configured in the account | R |
list_locations |
Locations configured in the account | R |
list_appointments |
Appointments by client/date | R |
get_appointment |
Single appointment | R |
create_appointment |
Book appointment | W |
update_appointment |
Reschedule/update | W |
list_intake_forms |
Submitted intake forms | R |
get_form |
Full form with answers | R |
list_questionnaire_templates |
Available form templates | R |
send_intake_form |
Send form to client | W |
list_notes |
Treatment notes (read-only in IntakeQ API) | R |
get_note |
Full note | R |
list_invoices |
Client invoices | R |
get_invoice |
Single invoice | R |
list_files |
Client file metadata | R |
audit_export |
Export HIPAA audit trail | R |
Tip: call
list_practitioners,list_services, andlist_locationsbeforecreate_appointmentto obtain valid IDs required by that tool.
MCP Resources
| Resource URI | Description |
|---|---|
intakeq://compliance-notice |
HIPAA compliance notice — BAA requirements and what the connector logs |
intakeq://auth-status |
Live authentication status (JSON) |
HIPAA Audit Log
Every PHI read and write is appended to ~/.intakeq-mcp/audit.log (JSONL format).
Each entry records: timestamp, session ID, machine IP, tool name, arguments (sanitized),
outcome, practitioner ID, client ID, and result count.
Export the audit log at any time: ask Claude to run audit_export.
Audit logging is fail-open. If the log file can't be written (disk full, permissions),
the tool call still completes rather than blocking clinical work — the failure is written
to stderr with an [audit-log-failure] prefix so it can be monitored/alerted on. If you
need a hard guarantee that PHI access is never left unlogged, monitor process stderr for
that prefix in your deployment.
HTTP transport
The server runs in stdio mode by default. To run as an HTTP server (e.g. for multi-user or remote deployments):
MCP_API_KEY=your-secret TRANSPORT=http PORT=3000 intakeq-mcp
The MCP endpoint is at /mcp, gated by a required Bearer token (MCP_API_KEY). The
server refuses to start in HTTP mode without MCP_API_KEY set, since an unauthenticated
endpoint would expose every PHI tool to anyone who can reach the port. To explicitly opt
into an unauthenticated endpoint for local development only, set MCP_ALLOW_NO_AUTH=true —
never do this with real patient data.
HTTP transport does not terminate TLS itself. Run it behind a reverse proxy (nginx, Caddy, your cloud provider's load balancer, etc.) that terminates HTTPS — otherwise the Bearer token and all PHI in transit are sent in plaintext on the network.
Each HTTP session maintains its own API key in memory — the set_api_key tool scopes the key to the session and never writes it to the shared disk file. Sessions are garbage-collected after 24 hours.
Health check: GET /health → { "ok": true, "sessions": <count> }
Environment variables
| Variable | Default | Description |
|---|---|---|
INTAKEQ_API_KEY |
— | Fallback API key for CI / headless use (prefer the set_api_key tool) |
INTAKEQ_API_BASE |
https://intakeq.com/api/v1 |
Override the IntakeQ API base URL |
TRANSPORT |
stdio |
Transport mode: stdio or http |
PORT |
3000 |
HTTP server port (HTTP mode only) |
MCP_BASE_URL |
http://127.0.0.1:3000 |
Base URL printed in HTTP startup logs |
MCP_API_KEY |
— | Bearer token to gate the HTTP /mcp endpoint. Required in HTTP mode unless MCP_ALLOW_NO_AUTH=true |
MCP_ALLOW_NO_AUTH |
— | Set to true to explicitly allow HTTP mode without MCP_API_KEY (local dev only) |
Rate limiting
IntakeQ enforces ~10 requests/minute. The connector retries automatically on HTTP 429 (up to 3 attempts, with back-off delays of 10 s / 20 s / 30 s). After 3 failures the error is surfaced rather than hanging indefinitely.
Not yet covered
- Insurance claims / CMS-1500 write
- Webhooks
- Bulk operations
- Creating treatment notes (not available in IntakeQ API)
Other open-source connectors by Oktopeak
Same architecture (encrypted key storage, audit logging, rate limiting), different platforms:
- Clio MCP — Claude + Clio practice management (
@oktopeak/clio-mcp) - MyCase MCP — Claude + MyCase legal practice management (
@oktopeak/mycase-mcp) - Filevine MCP — Claude + Filevine practice management (
@oktopeak/filevine-mcp)
Supporting this project
This connector is free, MIT licensed, and maintained by Oktopeak. It always will be — we don't take donations. If it saved you time, the things that actually help:
- Star this repo. It is genuinely how other firms find it.
- Tell another firm running IntakeQ.
- Leave a review if we helped you directly.
- Need it deployed, extended, or maintained for your firm? Commercial support — that is what funds the free work.
Who we are
Oktopeak — digital transformation for healthcare and law firms.
We're a 7-person in-house product team building AI solutions for regulated industries: AI integrations, workflow automation, and custom software our clients own outright. We maintain four open-source MCP connectors — Clio, MyCase, Filevine, and IntakeQ — and deploy them inside real practices with scoped credentials, audit logs, and workflows built around how your team actually works.
- 🌐 oktopeak.com
- 📅 Book a 30-min call
- ✉️ office@oktopeak.com — security reports welcome
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。