Skip to main content
The Superthread MCP server connects an AI client to your workspace. Add https://mcp.superthread.com/mcp as a remote MCP server in your client’s connection settings and follow its sign-in flow. Operations use the authenticated caller’s Superthread permissions. For built-in and custom Superthread agents, see agents. For terminal access, see the CLI reference.

​
Supported clients

The server page has step-by-step setup for:
  • Claude (web and desktop apps)
  • ChatGPT, through Developer mode on the web
  • Claude Code
  • Cursor
  • VS Code (GitHub Copilot agent mode)
  • Codex, including the ChatGPT desktop app, which shares Codex’s connection settings
  • Gemini CLI
  • Antigravity
  • Devin Desktop (formerly Windsurf)
  • Zed
Other clients that support remote MCP servers can usually connect with just the server URL.

​
Signing in

Most clients only need the server URL. They open a browser window where you sign in to Superthread and approve access. There’s no app to register, and no client ID or secret to enter. Claude, ChatGPT and VS Code identify themselves with a Client ID Metadata Document: a small file each client publishes that says who it is and where to send you back after sign-in. Superthread reads that file instead of asking you to set up an app. Other clients register themselves automatically the first time they connect. If a client can’t sign in through a browser, or you’re setting up automation, you can send a personal access token as a bearer token instead. The server page shows how for each client that supports it.

​
What your AI client can do

Once connected, your AI client can work with:
  • Cards, boards and sprints: find, create, update, archive and delete cards, with checklists, comments, tags and related cards. Build boards with your own columns, and manage sprints and sprint settings.
  • Projects and spaces: create and update projects, post health updates, and manage spaces and their members.
  • Pages: create, update, move, archive and duplicate pages.
  • Notes and templates: read and create notes, enhance or revise them, turn PII redaction on or off for a note, and create, edit and delete your saved note templates. See notes and templates below.
  • Tags: manage workspace and space tags.
  • Customers: manage customers, their tiers and statuses, and customer requests on cards and projects.
  • Agents: create and update custom agents, including their web access, triggers and sharing; give an agent access to your notes; and read or change your own instructions for the Superthread agent.
  • Favourites and sidebar: list, add, remove and reorder your favourites, and change which items and sections your own sidebar shows.
  • Forms and files: create, read and submit forms, and get links to files on cards and pages.
  • Your settings: your PII redaction defaults for note transcription and dictation.
  • Search: keyword, semantic and hybrid search across the workspace. See search below.
The server page lists every tool with a description. Tools for connected mail and calendar accounts appear there too, but they only work inside a custom agent’s run in Superthread, not from an MCP client. To have an agent work with Gmail, Outlook or Google Calendar, connect the account in the agent’s settings. See agents. The search tool takes an optional mode:
  • bm25 (the default) matches keywords. It suits titles, names and IDs.
  • semantic matches by meaning. It suits concepts and questions, like “what did we decide about billing retries”.
  • hybrid combines both, for when the client isn’t sure which fits.
Semantic and hybrid results include a score from 0 to 100 for how closely each result matches the meaning of the query. A workspace that can’t serve a mode rejects the request, and the client should fall back to bm25.

​
Notes and templates

Ask your AI client to read the note and its current enhanced version before editing it. Notes use exact saved note IDs, and templates use exact IDs returned by the template list. Titles and URLs are not substitutes for these IDs.

​
Note template operations

  • note_template_list lists all templates accessible to the caller, including their IDs, context, and ordered sections.
  • note_template_get reads a saved template by template_id.
  • note_template_create creates a template owned by the caller, with a template object containing title and optional context, sections, and is_shared. A new template is private unless is_shared is true. Creating a template doesn’t apply it to any note.
  • note_template_update edits an owned template by template_id, with an update object containing title, context, or sections.
  • note_template_delete permanently deletes an owned template by template_id. Notes already enhanced with it keep their enhanced content.
Omitted fields stay unchanged. context: "" clears the template instructions. sections: [] clears sections; any other sections array replaces the whole ordered list. Each section has title and description fields. Read the template before replacing its sections so you preserve sections you still need. Only the owner can update or delete a template. Shared and built-in templates may be readable without being editable. See note templates and the template API references.

​
Note enhancement operations

  • note_get reads the saved note, its revision, its enhanced entries in ai_notes, and its current last_note_template_id.
  • note_enhance starts a revision or re-enhancement using note_id and an enhancement object.
  • note_enhancement_get reads the latest operation for note_id, or a specific operation when you supply operation_id.
In enhancement, provide non-empty instructions, an explicit note_template_id, or both. Instructions alone revise the existing saved enhanced HTML. Optional enhanced_note_id chooses a saved enhanced entry; otherwise the entry matching last_note_template_id is used. An explicit template regenerates from raw user notes and the transcript, and can include extra instructions. Neither mode changes the original notes, transcript, or saved template. Supply the expected_revision from note_get to protect concurrent edits. An optional UUID operation_id identifies an identical retry. If omitted, the adapter supplies the current revision and an operation UUID. Reusing an operation ID with different input, starting a second operation while one runs, or using a stale revision fails with a conflict. Operations have idle, running, completed, or failed status. A running operation’s content is an unsaved preview. Only completed with ai_note confirms the enhanced result has been saved. If generation is still running, use note_enhancement_get with its operation ID. A failed operation is reported as an error: read the note again before retrying a conflict, and do not treat preview content as a saved result. Only the note owner can start or observe an enhancement. A custom agent’s note read grant permits note_get, but does not grant either enhancement operation. See enhance or revise a note, observe enhancement progress, and the CLI equivalents.