Integrating anonym.legal with Claude Enterprise

anonym.legal exposes an MCP (Model Context Protocol) server that detects, tokenizes, and detokenizes personal data (PII) in text. Connected to Claude Enterprise, it lets Claude redact PII before sending to third parties, restore original values on demand, and track cost/quota.

Who does what

• AdminRegisters the connector for the whole workspace, sets the organisation-wide permission ceiling, and controls the Cowork gate. Nothing here affects a specific user's credentials.
• UserAuthorises the connector for their own Claude account via OAuth and picks a per-tool preference (Always allow / Ask / Blocked) within the admin ceiling.

What you need

• A Claude Enterprise workspace (this guide applies to any Enterprise tenant).
• Admin role in Claude for Part A.
• An anonym.legal account for the OAuth handshake (users see a login window during connect).
• MCP server URL: https://anonym.legal/mcp.

Claude ships a directory of first-party and partner connectors, but anonym.legal is added as a custom Web connector. Start from https://claude.ai/admin-settings/connectors.

1. 1

   Click Add connector. The menu shows two choices: All available (curated directory) and Custom.

   

2. 2

   Hover Custom; a submenu offers Web or Desktop. anonym.legal is a hosted MCP endpoint, so choose Web.

   

3. 3

   The Add custom connector dialog opens (currently in BETA).

   

4. 4

   Fill in the fields:

   • Name: anonym.legal (displayed to users in chats and settings).
   • Remote MCP server URL: https://anonym.legal/mcp.
   • Leave OAuth Client ID and Client Secret empty. anonym.legal supports Dynamic Client Registration (DCR), so Claude negotiates client credentials automatically. Filling these fields is only necessary when the MCP provider does not support DCR.
   • Keep Individual sign-in selected (default). Each user authorises with their own anonym.legal account.
   • The Managed authorization (Beta) option is disabled unless your tenant is on the enterprise-managed-auth waitlist; managed mode would let your organization own the tokens centrally instead of each user.

   

5. 5

   Click Add. The connector now appears in the Connectors list. Users cannot use it yet — they each have to run through Part B.

   

Click the anonym.legal row to open its details. Admin permission choices form a ceiling: individual users can be equal or stricter, never more permissive.

Default restriction for unconfigured tools

The first control decides what happens when anonym.legal introduces a new tool that isn't yet listed here. Choose Blocked for the safest posture; existing configured tools are unaffected.

Read-only tools

These call anonym.legal without mutating state on the provider side. In this rollout all five are set to Always allow so end users can call them frictionlessly.

• Analyze Text for PII — detect PII entities in a text sample.
• Detokenize Text — reverse anonymisation using a session token.
• Estimate Operation Cost — dry-run pricing before real calls.
• Get Token Balance — remaining quota for the account.
• List Tokenization Sessions — sessions created by this user.

Write / delete tools

These tools create or destroy state on anonym.legal (a tokenisation session or its records). Your organization allows them, but sensitive tenants can force Approval required so Claude has to prompt the user before each call.

• Anonymize Text — replace PII with reversible tokens (creates a tokenisation session).
• Delete Tokenization Session — irreversibly drop a session and its tokens.

Independently of the per-tool table, Claude's Cowork/agent settings decide whether the "Always allow" tier is available at all for connector tools. Navigate to https://claude.ai/admin-settings/cowork.

• Enable for your organization — must be on for connectors to run agentically at all.
• Enable Dispatch — governs the Dispatch feature; unrelated to connector permissions.
• Allow "Skip all approvals" mode (Beta) — dangerous: users get a one-shot toggle to bypass every prompt. Leave off for regulated workloads.
• Allow "Always allow" for connector tools — when off, users cannot persist an "Always allow" preference; they get an "Ask each time" experience even for read-only tools. When on, users are still capped by the per-tool admin ceiling.

Use these steps when the connector needs a clean re-install (for example after changing the transport URL, rotating client credentials, or retiring the integration). Removal is organisation-wide and invalidates every user's OAuth grant.

1. 1

   Open the admin Connectors table at https://claude.ai/admin-settings/connectors. Locate the anonym.legal row.

   

2. 2

   Click the More options (⋯) icon on the row and choose Delete.

   

3. 3

   Confirm in the Remove anonym.legal for your team? dialog.

   

4. 4

   The table refreshes without the connector. Users who had it connected will see it disappear from their Customize page.

   

Once the admin has published the connector (Part A), each user activates it once on their own account.

1. 1

   Open the sidebar → Customize, or go to https://claude.ai/customize/connectors. The user-level Connectors panel lists everything the organisation has published.

   

2. 2

   If the row shows Get started, click it. A pop-up opens the anonym.legal OAuth authorisation page. Sign in with your anonym.legal credentials, grant access → the pop-up closes and the row flips to Connected. Tokens are stored per user; the admin never sees your credentials.

3. 3

   Click the row (or its View action) to open the connector detail panel. Here you can copy the MCP server URL, Disconnect, or reach the row action menu.

   

4. 4

   The More options menu offers Refresh tools list (re-fetch the tool catalogue from the server) and Remove (revoke only this user's grant — it does not delete the connector for the workspace).

   

Each row in Tool permissions shows a three-way radio: Always allow / Needs approval / Blocked. Those three are the user's own choices, bounded by the admin ceiling from §3.

Open a new chat and instruct Claude to invoke a specific anonym.legal tool. Example prompt used in this guide:

Use the anonym.legal tool "Analyze Text for PII" on this test input:
"Contact John Doe at john.doe@example.com or +49 30 12345678, IBAN DE89370400440532013000."
Return only the raw tool JSON result.

Claude invokes the tool and streams the JSON response. The screenshot shows nine detected entities including PERSON, EMAIL_ADDRESS, PHONE_NUMBER, and IBAN_CODE.

Admin vs user — what each side controls

Ceiling matrix

Individual sign-in vs Managed authorization

Security notes

• Only add connectors from developers you trust. A malicious MCP server can inject prompts through tool responses to trick Claude into taking harmful actions ("indirect prompt injection").
• The Default restriction for unconfigured tools setting is your safety net if the provider ships new capabilities without notice — set it to Blocked to require an admin review each time.
• Cowork "Skip all approvals" turns off every confirmation prompt in agentic runs. Do not enable it for tenants processing regulated data.
• Rotate credentials that were used to test connectors before promoting them to production users.

Where each setting lives