Answer your SayThing notes with your own agent (Claude Code, or any agent that speaks MCP and reads skills) instead of SayThing's AI. Your agent waits for a note, edits the model with checks and renders on our servers, and replies; the chat looks the same whoever answered, and turns it answers cost no credits (except generating a mesh, which is paid in credits whoever asks for it).
Two pieces, packaged as one Claude Code plugin: SayThing's MCP server at https://api.saything.app/mcp, signed in with your SayThing account over OAuth, and the skill: the loop (wait for a note, act, reply), how to act on a note, and the modelling rules, generated from the same files SayThing's own AI is prompted with.
/plugin marketplace add <the published git URL of the plugin>
/plugin install saything@saythingRestart, run /mcp, pick plugin:saything:saything and sign in in the browser. Then say "answer my SayThing notes" or run /saything.
claude mcp add --transport http saything https://api.saything.app/mcpthen /mcp → saything → authenticate, and give your agent the skill (skills/saything/SKILL.md from the plugin) as instructions.
A client that can't sign in can use a bearer token from your account: claude mcp add --transport http saything https://api.saything.app/mcp --header "Authorization: Bearer <token>".
Codex, Cursor and Claude Desktop add the same URL as a remote MCP server (Streamable HTTP) and sign in the same way. Claude Desktop and claude.ai can use it as a connector to design in the chat without the wait loop. Tried at launch: Claude Code; the others are untested.
Notes reach your agent only from objects set to "My own agent" (on the object's page here, or on the phone). A note on such an object waits until your agent is running; after ten minutes SayThing offers to send it to its own AI for credits instead.
wait_for_note, list_objects, open_object, create_object, read_model, write_model, edit_model, render, set_values, generate_mesh, ask, reply, release_note.