Connect your tools once and control what agents can reach

An account on another system, such as GitHub or Google Workspace, authorised once so that Cognibl's gateway can reach it on an agent's behalf. Connecting stores the credential, encrypted. It grants nothing on its own: each project then chooses which toolsets its agents may use.

94 providers today. Six are certified by Cognibl: GitHub, Google Workspace, Atlassian for Jira and Confluence, Linear, Notion and Microsoft Learn. 88 more are community servers from a curated list, including HubSpot, Stripe, Sentry, Vercel, Supabase, Box and Canva. From the store you can also search the official MCP registry.

Only where you allow it, and today only in Google Workspace. Read is the default for every tool. When an owner or admin connects Google Workspace, they can tick Create or Full for Gmail, Drive, Docs and Sheets, and Full for Calendar. Create lets agents make new things, such as a draft or a new file. Full also lets them change, send, share and delete what already exists. A project then has to turn each level on. These writes go straight to Google with nothing held for approval, so a wrong send is sent. Every call is traced. Every other tool is read-only.

No. An agent authenticates to Cognibl with a key Cognibl created, and only its hash is stored. The gateway holds the credential for the tool and makes the call. The agent receives the result, never the token that fetched it.

An admin has to stock it in the organization's store before anyone can connect it, and it is narrowed to the tools that declare themselves read-only, both when tools are listed and again when one is called.

Yes. Every call through the gateway, whether allowed, refused or failed, is written to the project's trace: append-only, hash-chained, and redacted at the moment of capture. The trace is written by the gateway, not reported by the agent.

Yes. Create a key on your account page, under Agent keys, and give it a label. The console hands you a ready mcpServers block for Claude Desktop, Claude Code or Cursor. The key acts as you. It can do what you can do in the console, under the same rules, and it reaches a project's tools only where that project has turned them on. Every call is recorded against you. A key expires after 90 days and you can revoke it at any time.

Entirely. Cognibl does not write connectors. Every system is reached through an existing MCP server, and adding one is configuration and policy rather than code. From an agent's side, Cognibl is one MCP endpoint and a key.