# Search as a named user without surfacing what they can't open

> Search as a named user without surfacing what they can't open: 0 companies fully do it, 16 with a limit, 37 carry the tag. Every listing in Company knowledge & data catalogues with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/company-knowledge.permission-scoped-search/
- Markdown: https://www.anchorterminal.com/niches/company-knowledge.permission-scoped-search/index.md (~2,000 tokens)
- Slim: https://www.anchorterminal.com/niches/company-knowledge.permission-scoped-search/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/company-knowledge.permission-scoped-search/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 0 | 16 | 37 | 20 | 9 |

## What counts as done

Results and answers are filtered to the user's access on every interface (REST, MCP, chat, AI-context helpers), and a user's tool list shows only tools their credential can call.

Measured, for our own tests: for a test set of restricted documents, zero appear in results or answers for a user without access.

How agents ask for it: search our docs as me; only show what this user is allowed to see; can the agent answer HR questions without leaking salaries

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Alation (Alation, Inc.)**: The remote MCP server lists tools per user and filters out those the user can't use, and user-account auth exists, but service accounts are admin-only, results follow Alation roles rather than source ACLs, and no per-interface test is documented.
- **Atlan (Atlan Pte. Ltd.)**: OAuth calls run as the signed-in user under existing Atlan policies, and Atlan says no tool exposes what that identity can't access, but that is the vendor's statement with no per-interface test, and these are catalogue policies, not source ACLs.
- **Basedash MCP (BaseDash Inc. (Delaware, United States))**: Each person connects by OAuth with scoped access (mcp:chat:ask, mcp:dashboards:read or write), and tools follow data-source access control and row-level security, but that rests on the vendor's statement with no per-interface test, and it covers data rather than documents.
- **ClicData MCP (ClicData SAS)**: The MCP page says every query runs inside the person's existing ClicData permissions, but that is the vendor's claim; tool definitions and auth detail need a signed-in account and the help article renders only with JavaScript.
- **Collibra (Collibra Inc.)**: The hosted MCP server signs each user in by OAuth and the docs say it respects that account's Collibra permissions, but no per-interface statement or read-only scope is given, and the local server uses one username and password.
- **DataHub (Acryl Data, Inc. (d/b/a DataHub))**: Cloud OAuth tokens are scoped to the signed-in user and Search Access Controls filter at query time, but Core only masks after search, these are DataHub's own view policies rather than source ACLs, and lineage filtering isn't documented.
- **Dust (Permutation Labs)**: The MCP client works as the signed-in user with MCP-scoped tokens and sees only that user's Spaces. Access is set by admins placing synced content in open or restricted Spaces, not by mirroring each source's ACLs, so a restricted file in an open Space is visible.
- **Gemini Enterprise (Google LLC)**: Ingested stores map users to ACL groups through identity syncs, and federation retrieves as the user. But identity sync runs on its own schedule (30 minutes at best), incremental syncs skip identity data, and no page states enforcement per interface or has a probe.
- **Glean (Glean Technologies, Inc.)**: Docs say every MCP tool runs as the signed-in user under connector ACLs, with user-scoped tokens. But a global token can act as anyone via X-Glean-ActAs, admin search ignores permissions, the ActAs path failed in acc-34, and no two-user probe exists.
- **Guru (Guru Technologies, Inc.)**: User tokens return what that user may see and custom sources can mirror per-record native permissions, but user tokens carry full write rights, and the read-only collection token gets HTTP 500 from the MCP server while REST accepts it.
- **Hexis Cloud (Bevelites GmbH)**: Per-path read rules in access.md apply on every branch, a folder no rule opens is invisible, and the tool manual is filtered per caller. No test of zero leaks is published, and v0.30.0 and v0.33.0 fixed access flaws without advisories.
- **Marmot (Marmot Data Ltd)**: Query tools are registered only when the caller may query, and writes check a manage permission, but search visibility per user isn't documented, there is no source ACL mirroring, and external OIDC tokens aren't accepted (issue 84).
- **Onyx (DanswerAI, Inc.)**: OAuth gives each user a read:search token under their own document access, but source permission sync covers 16 connectors on Business, Enterprise and Cloud only, one oversized mailbox aborts the sync (acc-19), and plugin tools are listed from admin credentials (acc-18).
- **OpenMetadata (Collate, Inc.)**: OAuth and personal tokens carry the user's own role, but AI Context lineage ignores the caller's permissions (acc-31), MCP ignores X-Impersonate-User where REST applies it (issue 37764), and service connections leaked unmasked on MCP until a fix on main (acc-30).
- **Overclock (Overclock Technologies Limited)**: Each OAuth connection is one user and one workspace, and fact tools withhold facts resting on documents 'you cannot see', but no page says how source ACLs are mirrored or how document search is filtered per user.
- **Secoda (Secoda, Inc.)**: The server sends one API_TOKEN for every call, and the README says an admin generates it, so an agent searches as that token's user with no scope; per-user filtering on Secoda's side couldn't be read.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Cohere North (Cohere Inc.)**: API and permission docs are closed by robots.txt. The North 2 post claims IAM integration and admin roles; the MCP SDK passes each user's identity and connector tokens to tool servers, but that is North as a client, not search filtering.

- not this job: Precisely AI Studio (Precisely Incorporated): Trial assets run on Precisely's sample data, and the page describes no search that filters by a named user's source permissions.
- not this job: Procore AI Go (Procore Technologies, Inc.): The app follows each user's Procore permissions and will not surface an RFI the user can't view, but only inside the iOS app; no agent path exists to search as a named user.
- not this job: Lens (Thought Industries) (Thought Industries, Inc.): Access is split by source groups chosen per widget, MCP connection or API key, not by each source's ACLs for a named user; Lens is built for customer-facing content.

## What stops agents finishing

- permission sync only for a handful of connectors or only on the enterprise licence (Onyx, per its founder)
- one oversized source aborts the permission sync
- only a global token that can act as anyone, or a scoped token the MCP server rejects

## Around it

Before: identity.user-token. After: company-knowledge.cited-answer.

## Where this comes from

Weight 20 of 100 in Company knowledge & data catalogues: 9 of 38 counted accounts (Onyx permission sync and per-user tool issues, OpenMetadata lineage and secret masking on MCP paths, Glean act-as and Guru collection token, Airweave thread), plus 1 second-hand podcast relay

As data: https://www.anchorterminal.com/api/v1/use-cases/company-knowledge.permission-scoped-search.json, or the MCP tool get_use_case with id company-knowledge.permission-scoped-search. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
