Connect Claude, ChatGPT or any MCP-compatible client to your organisation’s SCRS-governed knowledge. Retrieval stays inside the permissions you configure, personal data is replaced with placeholder tokens before anything leaves us, and every governed release is written to a tamper-evident audit log.
Opens Claude with the connector details filled in. Works on every Claude plan. For other clients, use the server URL below.
https://otherme.pophastalabs.com/scrs/mcpThe same URL for every customer — who you are, and what you are allowed to see, is resolved per request from your own access token. Add it as a custom / remote MCP connector in your client; there is no API key to copy, because the client registers itself over OAuth.
SCRS is a workspace product, so three things must be true before a connection will work:
If any of these is missing you will see an explanatory message at the approval step naming what to fix — nothing is shared and nothing is charged.
OAuth 2.1, discovered automatically. An unauthenticated request returns 401 with a
WWW-Authenticate header naming our protected-resource metadata, and the client takes
it from there.
S256, required.2025-06-18 (others negotiated on request).GET https://otherme.pophastalabs.com/.well-known/oauth-protected-resource/scrs/mcp
GET https://otherme.pophastalabs.com/.well-known/oauth-authorization-server| Tool | What it does |
|---|---|
ask_scrs |
Answer a question using this organisation’s private, SCRS-governed knowledge. This is the right tool whenever the answer depends on the organisation itself — its documents, policies, contracts, customers, people, products, pricing, processes, history or past decisions — or when an internal answer could differ from a general one; none of that information is in general training data. Personal data is pseudonymised before it reaches any model, so answers may contain bracketed placeholder tokens (e.g. [PERSON_1]) in place of real values. The same token refers to the same entity throughout, and the real values stay inside SCRS. |
searchread-only |
Search this organisation’s SCRS-governed knowledge base and return matching passages (redacted). The right tool when you need the organisation’s own source passages rather than a synthesised answer — quoting, checking exact wording, or gathering evidence. Returns a list of results with ids; call fetch(id) for the full passage. |
fetchread-only |
Fetch the full (redacted) text of a governed passage by the id returned from a previous search. |
list_documentsread-only |
List this organisation’s SCRS-governed documents that you’re allowed to see (file names are redacted). Use the returned id with read_document. |
read_documentread-only |
Read the full (redacted) text of one governed document by its id from list_documents. Personal data is pseudonymised before you receive it. |
ingest_document |
Add a file to this organisation’s SCRS-governed knowledge base: it is scanned, its text extracted, personal data pseudonymised, and stored — then searchable via search/ask_scrs. IMPORTANT: SCRS can only govern a file it receives FIRST. A file you were given in this chat has already been seen here; ingesting it governs the stored copy and future answers, not that first moment. Files a user sends to SCRS directly (or a link SCRS fetches) are governed end-to-end. |
An assistant decides for itself whether a question needs a tool, so naming SCRS is the reliable way to know an answer came from your governed knowledge rather than the model’s general training. Some examples that work:
To make this the default for a whole workspace, most assistants accept standing instructions — a Claude Project, a ChatGPT custom instruction, or organisation-wide instructions on a Claude Team or Enterprise plan. That makes SCRS much more likely to be used; it does not make it certain, for the reason set out below.
Expected, and the most common surprise. The assistant chooses whether to call a tool, so a question it believes it can answer from general knowledge often will be. Name SCRS in the question, or set a standing instruction. There is no setting — on our side or the assistant’s — that forces every message through SCRS.
An administrator has not switched the connector on for your workspace yet, or the workspace setup is incomplete. See Before you connect above. Nothing was shared.
Usually no model has been selected for the connector. An administrator picks one under AI module → Connect to Claude / ChatGPT.
[PERSON_1] or similar.That is the product working. Detected personal identifiers are replaced with stable placeholder tokens before any text leaves SCRS, so the assistant never receives the real values. The same token means the same person throughout an answer. An administrator can opt a workspace out of this for the connector, which is a deliberate and logged choice.
Access is resolved per person on every request, so it ends immediately if your workspace membership is deactivated, if an administrator revokes connections, or if the workspace’s plan or allowance lapses. Reconnecting will not bypass any of those.
Email support@pophastalabs.com. For anything security-related, security@pophastalabs.com.
What this does and does not cover. Connecting SCRS to a third-party assistant does not put SCRS in front of that assistant. It governs what SCRS returns. It cannot govern what someone types straight into Claude or ChatGPT — that goes to the assistant’s own provider and never reaches us — and it cannot make the assistant call us, because that is the assistant’s decision on every message.
For a guarantee that every request is governed, use a surface we control: the SCRS API, the embeddable chat widget, or the Other Me application, where the security layer runs on every request and fails closed.