Your AI assistant (MCP)
Sentinel connects to Claude and to any assistant that supports MCP (Model Context Protocol). The assistant reads your watches, notes and alerts and answers citing the source articles; with a write-enabled key, it can also create watches and add sources. Available on the Firm plan.
What you can ask
- “What did my Energy watch pick up on nuclear this week?”
- “Draft a two-paragraph client update from this morning’s weekly note.”
- “Who handles public affairs at this company, and what do my watches say about it?”
- “Create a weekly watch on industrial hydrogen in Europe.”
- “Add this outlet’s feed and a Google News search on RTE to my Energy watch.”
Before you start
- Your workspace must be on the Firm plan, including complimentary Firm access.
- In Sentinel, account menu, AI assistants gives the address to copy, links to Claude and ChatGPT, and the list of connected assistants, which you can revoke at any time.
- Claude and ChatGPT connect through OAuth: no key to copy, you sign in to Sentinel and choose the workspace the assistant will see. An API key is still needed for Claude Code and your own agents.
Connect Claude or ChatGPT
- Claude: Customize, Connectors, Add, Custom connector. ChatGPT: Settings, Plugins, then the plus button and Add custom MCP server, OAuth authentication.
- Name: Sentinel. URL:
https://app.sentinelbriefing.com/api/mcp. - Click Connect, sign in to Sentinel, choose the workspace and authorise. Keep “Stay connected” ticked so you do not have to reconnect every hour.
Read access is always granted; editing watches only if you tick it, and only for workspace admins.
Connect Claude Code
One command, then Sentinel shows up in /mcp:
claude mcp add --transport http sentinel https://app.sentinelbriefing.com/api/v1/mcp \
--header "Authorization: Bearer sk_live_..."Other assistants
Any MCP client that accepts a remote server (HTTP transport) and an authentication header works: address https://app.sentinelbriefing.com/api/v1/mcp, header Authorization: Bearer followed by your key.
Tools
| Tools | What they do |
|---|---|
list_watches, get_watch | Your watches, their settings, sources and activity over the past week |
search_analyses, get_analysis | Analysed articles: search by text, actor, signal type or date, then the full analysis |
list_notes, get_note | Your daily, weekly and monthly notes and flash alerts, in full text |
search_organisations, get_organisation | Atlas: an organisation, its declared executives and public affairs leads, its consultancies, and the signals from your watches that mention it |
hatvp_registrations, hatvp_relations | French lobbying register (HATVP): new registrations, consultancy clients |
list_trigger_documents, get_trigger_document | Parliamentary and NGO reports read by Sentinel, and the organisations they name |
create_watch, update_watch, set_watch_active | Write: create a watch, add or remove keywords and entities, change the cadence, pause |
add_source, add_news_search, attach_source, detach_source, search_sources | Write: add a feed, a website, an X, Bluesky or Reddit account, a podcast, a Google News search, or a catalogue source |
Security and privacy
- Read by default: writing requires an authorised key, and the assistant asks for your approval before each change. It never touches note recipients or sharing, and never deletes a watch or a source.
- Isolated per workspace: a key only sees and changes the workspace it belongs to. The HATVP register, Atlas and public reports are shared, read-only.
- Traceable: every change is recorded in your workspace log, in the key holder’s name.
- Content flagged as external: responses are marked as coming from articles and reports, so the assistant never mistakes article text for an instruction.