frwrd.to

mcp for agents

Connect an agent such as Claude to frwrd.to over MCP, with your workspace key, and see which tools it gets.

MCP, the Model Context Protocol, is a standard way to give an AI agent tools it can call. frwrd.to offers one tool for each thing the API does, so an agent can make links, write pages and read numbers in your workspace.

The endpoint is https://api.frwrd.to/mcp. It speaks streamable HTTP, and it takes your workspace key as a bearer token:

Authorization: Bearer fw_...

An agent sees only the workspace its key belongs to.

Claude Code

claude mcp add --transport http frwrd https://api.frwrd.to/mcp --header "Authorization: Bearer fw_..."

Run claude mcp list to check the connection. Keep the key out of files you commit: if you share a .mcp.json, write the header as Bearer ${FRWRD_KEY} and set FRWRD_KEY in your environment.

ChatGPT and claude.ai

These apps cannot send a header, so they sign in with OAuth instead. In ChatGPT, turn on developer mode (Settings, Apps, Advanced), create an app with the URL https://api.frwrd.to/mcp and OAuth as the authentication, and connect. In claude.ai, add a custom connector with the same URL. Either one opens a frwrd.to page that asks whether the app may use your workspace: allow it, signed in or with your key. The app keeps access until you rotate your workspace key. An app connected this way cannot set or remove the recovery e-mail, claim a domain or delete the workspace: those need the key. Settings, Account, "Connected apps" lists every app connected this way and ends one with Revoke (also list_connected_apps and revoke_connected_app).

Other MCP clients

Add a remote server with the streamable HTTP transport (some clients call it "http"), the URL https://api.frwrd.to/mcp, and the header Authorization: Bearer fw_.... Clients that only speak stdio can reach it through a bridge such as mcp-remote.

The tools

Workspace:

  • get_workspace: the workspace, its limits and the DNS record that verifies the domain.
  • verify_workspace: check the DNS record now and lift the limits.
  • claim_handle: claim a short handle so pages live at /@handle.
  • set_utm: set the UTM defaults, appended to every link you create or retarget and to the links on your pages.
  • set_default_language: choose the language that lives at plain page addresses.
  • list_domains, add_domain, remove_domain: custom page domains.
  • list_connected_apps, revoke_connected_app: the apps connected through OAuth; revoking needs the key.
  • delete_workspace: delete the workspace; it needs confirm, the workspace's domain, and an agent should call it only when you asked for exactly that. Deleting a workspace stops its key and takes its pages offline at once; after 7 days the pages and settings are deleted. Short links keep working. Proving the domain (key recovery) within 7 days restores everything; up to 60 days, the links. After 60 days the domain is free and the links keep working without an owner. Browsers may keep cached copies of a page, its QR code and images for up to an hour after the delete.

Links:

  • create_link: a short link for an http or https address.
  • list_links, get_link: read your links.
  • update_link: change the target, title or tags, or archive and restore a link.
  • archive_link: stop a link from redirecting, keeping its numbers.
  • get_link_stats: clicks in a time window, by channel and by day; range (1d, 7d, 30d) adds every hourly or daily bucket and the all-time totals.
  • get_workspace_stats: the numbers of all your links at once: the clicks in a range (1d, 7d, 30d; default 7d) as buckets with the totals, the all-time total (null when it could not be counted right now) and the ten most clicked links. The answer is cached, so it may be up to 5 minutes old; when the server is busy it answers rate_limited, so retry in a few seconds.
  • get_qr: the link's QR code as SVG.

Pages:

  • list_pages: your pages, with language and translations.
  • get_page: a page as Markdown, with its version and the short link behind each link.
  • write_page: create or replace a whole page.
  • delete_page: remove a page.
  • get_page_stats: views of a page and clicks on each link; range works as above.
  • list_page_versions: the save history.

Each page is also a resource, frwrd://pages/{path}, with - for the root page. Tool errors carry the same codes as the REST API.