Give full context to your agents
Coding agents guess because they don't know what you know. Connect your Constella library over MCP and Claude Code, Codex, and every assistant work from your real memories — not generic web pulls.
Your library feeds every agent → they pull the context they need → grounded, actually-useful output.
Browser sign-in needs nothing copied. For a tool that cannot open a sign-in page, copy your key in the Constella app: Settings → Developer → Your key. Send X-Constella-Tools: catalog with it, or the tool will not see memory_add. Get Constella →
- Add Constella:claude mcp add --transport http --scope user \ constella https://api.constella.app/mcp
- Start
claude, type/mcp, pickconstellaand choose Authenticate. - Your browser opens. Sign in to Constella and press Allow.
- Test: “What do I know about X? Remember that Y.”
- No browser? Use your key instead:claude mcp add --transport http --scope user \ constella https://api.constella.app/mcp \ --header "Authorization: Bearer $CONSTELLA_API_KEY" \ --header "X-Constella-Tools: catalog" \ --header "X-Constella-Client: claude-code"
The Constella desktop app also runs its own MCP server on your computer, constella-canvas. Its Connect button adds it to Claude Code and Codex, and it lets an AI draw on your open canvas while the app is open. It is separate from the cloud address above: use either, or both.