MCP server

Typelets ships an MCP server, @typelets/mcp, so an MCP-capable client (an AI assistant or agent) can manage workspaces, files, and problems on your behalf.

Install

The server is published to npm as @typelets/mcp. Run it with npx and point your MCP client at it:

npx @typelets/mcp@latest

Use @latest (or pin a version) so you pick up new tools; an older cached version will not have them.

Authentication

The server authenticates to the Typelets API with a personal access token (PAT). Create one in your Typelets account settings and provide it to the server via its environment (for example TYPELETS_TOKEN). Treat the token like a password - it grants access to your workspaces.

What it can do

The server exposes tools across a few areas. The exact set depends on the installed version; this reflects the current release.

Workspaces

list_workspaces, get_workspace, create_workspace, delete_workspace
whoami - confirm which account the server is acting as

Files

list_workspace_files, read_workspace_file
create_file, update_file, append_to_file, delete_file
create_folder, delete_folder, move_path
upload_file - upload a binary file (image, font, PDF, asset). Bytes are base64-encoded; max 25 MiB. On a persistent workspace the asset is served at the preview URL / custom domain. Use create_file / update_file for UTF-8 text.

Problems (interview library)

list_problems, get_problem, save_problem_to_library, edit_problem, delete_problem, apply_problem_to_workspace

Sessions and review

list_recordings, summarize_recording, score_against_rubric, suggest_followup_questions
list_pending_invites

Hosting via MCP

You can build up a workspace’s files entirely through the file tools and serve the result with persistent preview. Text files go through create_file / update_file; images and other binaries go through upload_file. On a running persistent container, file changes are pushed into the live preview.