MCP tools
External tool servers behind the same compliance gate, with per-key grants that are denied by default, so agents only reach the tools you allow.
Register external MCP tool servers (HTTP, SSE, or stdio) in the Console with residency tags; auth secrets are sealed in the credential vault and never echoed back. MCP services follow the residency policy, not the signed DPA's model-provider list.

Every tool call clears the same gate as a chat completion: Inspect (data protection on the arguments), Route (residency on the server), Seal (audit chain), Meter (rate and budget). Tool grants are per-key and deny-by-default. Sluis also answers POST /v1/mcp as an MCP server itself, exposing exactly the tools the calling key is granted.
Call tools from your own code
Use POST /v1/mcp with an API key to see the tools that key may use. Running a tool is a separate call.
- List:
tools/listreturns the tools your key is granted. A server is left out when residency does not allow it, its sign-in is missing or its tool list cannot be fetched. - Ask:
tools/callruns nothing. It returnsneeds_confirmationwith the tool, the arguments and where to run it. - Run: send the tool's arguments as the request body to
POST /v1/mcp/servers/{id}/tools/{tool}/call. This call goes through Inspect, Route, Seal and Meter. - A tool your key is not granted fails with error
-32000, bad parameters with-32602and an unknown method with-32601. JSON-RPC errors come back with HTTP 200; a malformed request returns HTTP 400. - Stdio servers only run executables the operator has allow-listed in
SLUIS_MCP_STDIO_ALLOW, up to 16 at a time.
# list the tools this key is granted (deny-by-default) curl https://api.sluis.ai/v1/mcp \ -H "Authorization: Bearer $SLUIS_KEY" \ -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/list" }'
# ask for a granted tool: nothing runs, the answer says where to run it curl https://api.sluis.ai/v1/mcp \ -H "Authorization: Bearer $SLUIS_KEY" \ -d '{ "jsonrpc": "2.0", "id": 2, "method": "tools/call", "params": { "name": "crm.lookup", "arguments": { "email": "j.devries@acme.nl" } } }'
{
"jsonrpc": "2.0",
"id": 2,
"result": {
"needs_confirmation": true,
"tool": "crm.lookup",
"arguments": { "email": "j.devries@acme.nl" },
"note": "Execute via POST /v1/mcp/servers/{id}/tools/{tool}/call so the compliance gate (Inspect→Route→Seal→Meter) applies."
}
}# run it: the body is the tool's arguments; the call clears Inspect → Route → Seal → Meter curl https://api.sluis.ai/v1/mcp/servers/$SERVER_ID/tools/crm.lookup/call \ -H "Authorization: Bearer $SLUIS_KEY" \ -d '{ "email": "j.devries@acme.nl" }'
MCP authentication: two directions
Sign in to Sluis from an MCP client
Inbound OAuth at /agent/mcp signs an MCP client in to Sluis through the browser. No key is pasted into this setup, and the keys for model calls are unchanged.
- Add Sluis in your client, as in the examples below. The client finds Sluis through the published discovery documents and registers itself where it supports that.
- The client opens your browser on the consent page,
/console/mcp-authorize. It shows the client, the organisation and the tools it can reach. - Approve or deny access in the intended organisation.
- Sluis creates or reuses a personal tool credential for this client automatically.
- Existing plugin bindings, tool grants and policy still apply. Consent alone grants no tools.
- The organisation needs Agent Harness enabled.
- Access tokens last one hour. Refresh tokens rotate on every use.
- Revoke a client under Agent Harness › Connected clients. That disconnects its tokens and personal agent key.
- This is separate from upstream
user_oauthsign-in.
claude mcp add --transport http sluis https://<gateway>/agent/mcp
codex mcp add sluis --url https://<gateway>/agent/mcp codex mcp login sluis
Configuration examples, not a verified interoperability matrix. External Claude, Codex and upstream provider interoperability has not been established by these examples.
Static server credentials
Use headers for fixed headers. The legacy oauth value means a static bearer token stored in the vault, not browser sign-in.
Account mode
For interactive OAuth, account mode decides whose identity Sluis uses with the tool server.
- Organisation (
account_mode=organization): a shared identity that an admin connects explicitly with OAuth. It never reuses a personal credential. - Each user (
account_mode=user): only the acting member's own connection. - Authentication is separate: interactive OAuth uses
user_oauth, and the legacyoauthvalue stays a static bearer token.
Personal upstream connections
In Each user mode, every member signs in to the upstream service themselves.
- An admin grants the server to Workspace. Admin-added servers need this first.
- Each member signs in under Settings › Connections.
- A connection does not bypass tool grants, residency, data protection or budgets.
- Admins can view status and revoke connections, but cannot use another member's identity.
- Calls without a user actor, including organisation-owned keys, are refused in Each user mode.
- If you have not connected, or the service no longer accepts your sign-in, the tool returns
connect_requiredand the answer shows a Connect action for that server. Nothing runs under another identity.
Switch MCP tools off or on
Owners and admins switch MCP tools in Workspace conversations in the console's Workspace configuration. The same setting is the boolean mcp_enabled on GET/PUT /chat/settings.
- It is true by default: for new organisations, legacy settings and an absent field. Leaving it out of a
PUTkeeps the current value, including false. - When false, Workspace does not discover MCP tools and refuses to run them, including tools found earlier in existing conversations, Agents and subagents.
- Registrations, approvals, grants and credentials are kept.
- External
/agent/mcp,/v1/mcpand the console Playground are unaffected. - The personal server registration policy stays independent and organisation-wide.
Personal MCP suggestions
Personal MCP suggestions are Off by default. Admins choose one of two settings.
- Off pauses existing personal servers. It does not block sign-in to admin-added servers.
- Require admin approval lets members submit private, requester-only servers.
Each request goes through a review.
- A request names only the server and its endpoint. While pending, it neither contacts the server nor grants tools.
- Admins review the tools, residency and authentication explicitly. Approval never grants wildcard access or promotes a server to the organisation.
- Approval sets the reviewed configuration and an explicit list of tool names, available to the requester alone. The requester then signs in under Connections if the server uses OAuth.
- Private servers support
none(no upstream sign-in) or interactive OAuth, not personal static secrets. - Removing an approved request revokes the server and its connections.
Confirmations for changes to data
Workspace separates tools that read from tools that change data.
- A connector's own read-only label (
readOnlyHint) is used when it sends one. - A tool with no label counts as changing data, unless an admin marks it as read-only in the connector settings.
Before a changing tool runs
The member sees the connector, the tool and the exact arguments in a card and chooses Run or Don't run.
- No answer within five minutes, closing the chat or pressing Stop counts as declined, and the model is told the user declined.
- A scheduled agent run has nobody to ask, so there a connector that asks never runs a changing tool.
Admin settings per connector
- Never ask: recorded in the audit log and shown on the connector. Read-only tools never ask.
- Read-only tools only: the AI model is not offered any tool that changes data. Use it, for example, for a connector that uses a shared account.