QP-Insights MCP Connector

Connect your AI assistant to our QP-Insights platform. This is a standard remote Model Context Protocol (MCP) server, so it works with any MCP-compatible client that supports OAuth 2.1 — Claude, or another compliant assistant. Ask questions about your projects, subjects, exams and analysis results in natural language, and let your assistant manage your projects and read the data you already have access to in QP-Insights.

 

What this connector does

QP-Insights is a platform for managing and analyzing clinical imaging studies — DICOM data organized into projects, sites, subjects, exams and series, with quantitative analysis pipelines run over them and associated clinical data.

This connector gives your AI assistant a read-first window onto that platform, so you can:

  • Explore a project — how many subjects and exams it holds, how they split across sites, which analysis pipelines are available.
  • Find and inspect a subject — search by subject code, read demographics, exam counts, tags and the status of the latest analysis.
  • Browse imaging data — list a project’s or a subject’s exams, the DICOM series inside an exam, and the analyses that have been run on it.
  • Read analysis results — check whether a pipeline run has finished, then read its structured measurements.
  • Organize with tags — read tags on subjects, studies and series, and (with explicit permission) add to them or replace them outright.
  • Track queries — list and read queries raised against a subject or exam, and, with explicit permission, raise new ones, change their status or assignee, and comment on them.
  • Register a new subject — allocate a valid subject code and create the subject record (with explicit permission).
  • Record quality control — mark an exam or a subject as PASS or FAIL in quality control (with explicit permission). Each verdict overwrites the one previously stored.
  • Administer the platform — if you are a QP-Insights administrator, create and reconfigure projects and sites, and create, update, disable or re-invite users (with explicit permission).

 

What it deliberately does not do

  • It cannot launch analysis pipelines. Pipeline execution is triggered inside QP-Insights, by design. Your assistant can list pipeline definitions and read results but never start a new analysis.
  • It cannot read original patient data such as imaging DICOM files or clinical data filled through the electronic clinical report forms (eCRF).
  • It cannot create, edit or delete analysis pipelines.
  • It cannot delete any information.
  • It cannot upload or modify imaging data. DICOM ingestion stays in the platform.
  • It stores nothing. The connector holds no copy of your data.

 

Who can use it

You need an existing QP-Insights account. The connector cannot create an account for you — it authenticates as you — and it never widens your access: it can only ever see and do what your own QP-Insights account can.

If you do not have an account, contact your Quibim representative or your organization’s QP-Insights administrator.

Note for IT administrators: each QP-Insights deployment is served by its own MCP instance, bound to exactly one platform URL. If your organization runs a dedicated QP-Insights environment, ask Quibim for the MCP URL that matches it.

 

Setting it up

This is a standard remote MCP server with OAuth 2.1 dynamic client registration, so the setup is the same shape in any compliant client:

  1. In your client, find the option to add a custom/remote MCP connector or server (in Claude: Settings → Connectors → Add custom connector).
  2. Paste the MCP URL for your QP-Insights deployment.
  3. Your client discovers the connector’s authorization server and registers itself automatically. You will see a consent screen listing the permissions being requested — review it and approve.
  4. You are then redirected to Microsoft sign-in to authenticate with your organizational account. A Microsoft account chooser is shown so you can pick the right identity.
  5. Once sign-in completes, the connector is active and its tools appear in your client.

Your session is tied to your identity. Nobody else’s account can use your QP-Insights access, and disconnecting the connector in your client revokes it.

 

Example prompts

  • List my QP-Insights projects and tell me how many subjects each one has.
  • In project ONCO-2, how do the subjects split across sites and by sex?
  • Find subject SUBJ-0042 and summarize their exams — modality, date, and whether they’ve been segmented.
  • Which analysis pipelines are enabled for project PROSTATE-01.
  • What inputs does QP-Prostate need?
  • Has the analysis on this study finished? If so, can you summarize the results?
  • Get the analysis results for patient SUBJ-0042 across all timepoints
  • What open queries do I have across all my projects?
  • Open a query on subject SUBJ-0042 — the consent form is missing — and assign it to Marc Smith. (requires tickets:write)
  • Tag all exams from patient SUBJ-0042 as “Reviewed”, public. (requires tags:write)
  • Set up a new site — General Hospital in Spain, site code 004 — on project ONCO-2, then invite Marc Smith as a Coordinator there. (requires admin:read + admin:write)
  • Who has access to project ONCO-2, and with what roles? (requires admin:read)
  • Close all opened tickets in project ONCO-2. (requires tickets:write)
  • Set all exams from subject SUBJ-0042 as quality check pass. (requires quality:write)

Your assistant will typically resolve the project first, then the subject, then drill down — so giving it the project or subject code up front makes answers faster.

Note that to run some of the previous prompts specific roles are required, if your assigned role cannot perform the specific action the AI assistant will let you know.

 

Data handling

This connector is a pass-through. It does not store your data.

  • No persistence of platform data. Projects, subjects, exams, series and analysis results are fetched from QP-Insights on demand, returned to your assistant, and not written to any database or disk by the connector.
  • What is stored: only OAuth session state — registered client records, short-lived authorization codes, and access/refresh tokens. Access tokens live 15 minutes; refresh tokens live 30 days. No patient or imaging data is ever placed in that store.
  • Authorization stays with the platform. Your own token is forwarded to QP-Insights on every call. If you cannot see a project in the QP-Insights UI, the connector returns an access error, not the data. It cannot escalate your privileges.
  • Log redaction. Server logs strip bearer tokens and identifying fields — subject codes and code hashes, study and series instance UIDs, patient name and id, birth date, and eCRF payloads — from both URLs and request bodies before anything is written.
  • eCRFs are not available through the connector, only the completion status is returned.
  • Data leaves your control when you use it. Content the connector returns is sent to the AI assistant you’re using for processing, subject to that provider’s own data policies (for Claude, see Anthropic’s data policies). Treat this the same way you would treat any other export from QP-Insights, and follow your organizations’ policy on protected health information.

 

Clinical use

QP-Insights data is clinical imaging data. This connector is intended for authorized Quibim customers and internal research use. Output produced by your AI assistant from this data is not a medical device output and must not be used as the basis for a clinical decision without review in the QP-Insights platform itself.

 

Security

For IT administrators evaluating this connector:

ControlImplementation
TransportHTTPS only. TLS terminates at the ingress; HSTS enabled. Certificates managed in Azure Key Vault.
AuthenticationOAuth 2.1. The server acts as its own authorization server and delegates identity to Microsoft Entra ID. PKCE (S256) is mandatory. Grants: authorization_code and refresh_token only. Tokens are RS256-signed; the public key is published at /.well-known/jwks.json.
Client registrationDynamic (RFC 7591). No shared static client credentials are distributed.
AuthorizationPer-tool scope enforcement at the server, plus the upstream platform’s own permission model on every call. Read scopes are the default; write scopes are opt-in.
Origin validationRequests to the MCP endpoint carrying a browser Origin header are checked against a per-deployment allow-list (this deployment: https://claude.ai and its subdomains), mitigating DNS-rebinding. Requests without an Origin header — the normal case for desktop and CLI clients — are unaffected; a browser-based client outside the allow-list gets a 403.
Rate limitingPer-IP limits at the ingress and again in-process on every OAuth endpoint, with a tighter cap on dynamic client registration.
SecretsNever baked into container images or deployment charts. Sourced at runtime from Azure Key Vault via the Secrets Store CSI driver.
Runtime hardeningContainers run non-root with a read-only root filesystem, all Linux capabilities dropped, seccomp RuntimeDefault, and no service-account token mounted. Session state store is network-isolated to the application pods.
AvailabilityHorizontally autoscaled with a pod disruption budget; sessions are not pinned to an instance.

 

Troubleshooting

I’m asked to sign in repeatedly. Access tokens are short-lived by design and refresh silently. Repeated interactive prompts usually mean the refresh token expired (30 days of inactivity) or an administrator revoked the app grant. Remove and re-add the connector.

A tool says I don’t have permission. Two possibilities. Either the scope was not granted when you approved the connector — remove and re-add it, approving the permission you need — or your QP-Insights account genuinely lacks access to that project or subject, in which case an administrator has to grant it in the platform.

A browser-based client gets a 403 connecting to the MCP endpoint. The deployment’s Origin allow-list doesn’t include your client’s origin. Ask your QP-Insights administrator to add it. This doesn’t apply to desktop/CLI clients.

My assistant can’t find a subject I know exists. Subject codes are scoped to a project. Make sure your assistant is working in the right project, and ask it to search subjects with a partial code — the search matches substrings, case-insensitively.

An analysis has no results. Results only exist once the pipeline run has finished. Ask your assistant to check the analysis status first. Note that the connector cannot start a run for you.

Asking for a report PDF gets you nothing. There is no tool that downloads an analysis report; a report-generating analysis can be listed, but its PDF has to be opened in QP-Insights itself. Ask for the structured measurements instead.

 

Quibim Website
Privacy Overview

When you visit any website, it may store or retrieve information on your browser, mostly in the form of cookies. This information might be about you, your preferences or your device and is mainly used to make the site work as you expect it to. The information does not directly identify you but can give you a more personalized web experience. Because we respect your right to privacy, you can choose not to allow some types of cookies. Click on the different category headings to learn more and change our default settings. However, blocking some types of cookies may impact your experience of the site. You can find more information, including a detailed cookie explanation, on our Cookie Policy.