Tuteliq runs a hosted Model Context Protocol server at https://api.tuteliq.ai/mcp. Point an MCP client at it, sign in through the browser, and your assistant gains tools spanning detection, moderation, verification and compliance. Nothing to install, no API key to paste.
Connect a client
Claude Desktop: Settings → Connectors → Add custom connector, name it Tuteliq, URL https://api.tuteliq.ai/mcp. Claude Code, Cursor, Windsurf and other clients that support remote servers use this config:
On connect your browser opens a Tuteliq sign-in page. Approve, and the client stores the token. In Claude Code, run /mcp if the flow does not open on its own.
In Claude Code you can skip the config file and install the plugin instead with /plugin marketplace add Tuteliq/claude-plugin, which pulls from github.com/Tuteliq/claude-plugin. The server itself is documented at github.com/Tuteliq/mcp. See the Claude plugin page for the full install flow.
Authentication
OAuth 2.1, self-registering
Dynamic client registration (RFC 7591), PKCE with S256, and discovery via RFC 8414 and RFC 9728. No manual credential exchange.
30-day access tokens
Every token response carries expires_in and a refresh token valid for 90 days, so an idle client reconnects without user interaction.
Rotating refresh tokens
Each refresh spends the presented token. Replaying one revokes every token descended from that grant, deliberate, and one re-authorisation to recover.
Headless environments
OAuth needs a browser, so CI pipelines, cron jobs and containers use a static token generated in the dashboard under Settings → Plugins. Keep it out of version control, and prefer OAuth wherever a browser exists.
What it covers
Detection quality is identical to the REST API, the MCP server is a transport, not a different model. Both surfaces are backed by the same tool definitions.
Working a moderation queue
Size it
get_incidents_overview returns totals, requires_review_count, 24h/7d/30d counts and breakdowns by category, severity, source, status and platform. get_incident_trends buckets by hour, day or week.
Triage
list_incidents is the queue, filterable by status, severity, category, source, date range and your own identifiers. Every row carries recommended_actions, so an agent can rank without fetching each incident.
Inspect
get_incident returns the full record: risk category and level, confidence, detected_patterns, recommended_actions, modality, review state and summary.
Decide
review_incident records one decision, batch_review_incidents applies one decision across up to 100, and each still emits its own signed receipt.
What an agent may decide
An MCP token identifies an account, never an individual, so an agent cannot act with the authority of a named senior moderator. Escalation, resolution and reopening stay with a human.
Example prompts
Review this chat export for grooming risk and tell me what needs a moderator.
How many critical incidents came in this week, and where do they concentrate?
Work the moderation queue: triage anything at flag_for_review or above.
Pull a signed Article 12 audit receipt for that detection request.
No user content stored
Messages, images, audio, video and documents are processed in memory and never persisted, raw or derived.
Signed audit trail
Every moderator decision emits a signed EU AI Act Article 12 receipt, batch decisions included.
Annotated tools
Every tool carries MCP annotations, so a host can tell read-only from destructive before invoking anything.
Ready to connect?
A Tuteliq account is required. Start with a 14-day free trial, or install the Claude plugin for a one-command setup.