Connect AWS once for daily evidence, bring the agent into Slack, and reach your ISMS from the AI tools you already use. Each connection is scoped and read-only by default.
Evidence and context arrive without anyone copying them over.
01
Give Secani a read-only role in your AWS account. It checks root MFA, root access keys, Config recording and Security Hub findings every day.
The role is assumed with an External ID for 15 minutes; Secani refuses to run if the role works without it.
02
Ask Secani in Slack and it reads the thread, searches channels and answers with context from your ISMS.
Secani connects through Slack's official MCP server; posting a message on your behalf needs your approval.
03
Claude, ChatGPT, Cursor and Codex reach your ISMS through MCP: search controls, read evidence, propose changes.
Every write asks you in the AI client first; if the client cannot ask, nothing is written.
The copying between your tools and your ISMS.
Screenshotting the AWS console for MFA and Config evidence
Collected daily through a read-only role
Searching Slack for the decision behind a control
The agent reads the thread and cites it
Pasting ISMS context into an AI chat
Your AI client reads it directly through MCP
Retyping an AI suggestion into Secani
Proposed through MCP and saved after your confirmation
How often does someone copy evidence or context between tools?
Hours per year spent copying
42
Your numbers, multiplied: copies × minutes. Nothing of ours goes into the result.
Every connection can do what it needs and nothing more.
The AWS session is limited to four read actions, requires an External ID and checks that the account matches the connection.
Posting to Slack and writing through MCP both wait for your approval; read-only mode blocks writes for a client entirely.
Every MCP call and every AWS run checks the user's access again, scoped to organization, workspace and governance scope.
AWS evidence collection, the agent in Slack, and MCP for Claude, ChatGPT, Cursor and Codex. Microsoft Teams is prepared and needs setup on your side.
Only four read actions: the IAM account summary, the Config recorder status, Security Hub findings and the caller identity. It cannot change anything.
No. Reading threads and channels works directly; sending a message on your behalf needs your approval.
Yes. The OSCAL API validates OSCAL documents, with key limits of 120 requests per minute and 10,000 per day, and MCP gives authenticated access to your workspace.
Connect what you have and let evidence arrive on its own.
