src ▸ client linked 0

Give Claude Code real web actions on any site

Give Claude Code callable web actions on any site: one command to add the Writ MCP server.

Get started free See all MCP clients
Writ runs on your own accounts, with your own credentials and data, on sites you are authorized to use.
terminalstdio
stdio 
claude mcp add writ-cloud \
  -e WRIT_API_KEY=wt_your_api_key \
  -- npx -y writ-mcp
read ▸ what it is 01
01
what it is

Writ tools, inside the coding agent

Claude Code adds MCP servers from a single command. Connect Writ and your agent can drive real websites mid-session (submit a form, export a report, check a dashboard) and read the structured result back into the work.

  1. 1

    One claude mcp add command wires in every workflow you publish; scope a wt_ key per project so a session only sees the tools it should.

  2. 2

    The agent acts on the live site as you (not a read-only crawl) so it can complete tasks, not just summarize pages.

tools ▸ what they do 02
02
tools

What Claude Code can do with a Writ tool

Each tool is a published browser workflow - it acts on the live site, then returns structured output.

⊘

Authenticated actions

Claude Code acts on your own accounts - log in, pass TOTP and mailbox-OTP 2FA, then click, fill, and submit. Not a read-only crawl.

◍

Any site, no API required

Each tool is a recorded browser workflow, so it reaches sites that ship no API at all - internal dashboards, vendor portals, no-API web apps.

{ }

Typed, structured output

Every tool returns typed JSON Claude Code can read and act on - fields, not raw page text - so the agent can chain the result into the next step.

wire ▸ connect it 03
03
connect

Add Writ to Claude Code over stdio

Paste the server block, authenticate with a scoped wt_ key, and your published workflows appear as tools.

terminalstdio
stdio 
claude mcp add writ-cloud \
  -e WRIT_API_KEY=wt_your_api_key \
  -- npx -y writ-mcp
  1. 1
    Step 1

    Run the claude mcp add command; -e puts your wt_ key in the environment, not in the process list or your shell history.

  2. 2
    Step 2

    Confirm the server is connected with claude mcp list.

  3. 3
    Step 3

    Your Writ workflows are now callable tools in the session.

What this client gets One-command setupScoped wt_ keys per project
⊘

Why this setup is different. Claude Code is a terminal client, so this is the one setup that avoids a pasted HTTP stanza on purpose: a key passed as a --header flag would sit in argv — readable by any local process through ps and written verbatim into your shell history. The writ-mcp connector takes the key as an environment variable via -e instead, and registers as writ-cloud so it never clobbers the Writ desktop app’s own writ server.

{ }

The tool list is your workflow list. Anything you publish in Writ - a recorded flow, an AI-built workflow, a discovered API - shows up to Claude Code as a callable tool with a typed schema. See how MCP tools work or read the developer MCP docs.

auth ▸ logged-in web 04
04
the hard web

Login, 2FA, and dynamic pages - handled

The sites worth automating are usually the hard ones. Claude Code gets tools that handle them.

Because every tool runs the real browser flow, Claude Code can complete tasks a read-only MCP server never could: sign in, pass two-factor, navigate pages that render with JavaScript, and submit. It runs on your own accounts and credentials, on sites you are authorized to use.

PersonasTOTP & mailbox OTP AI navigationReal Chromium rendering 30+ step types
build ▸ the tool 05
05
build

Three ways to author a Claude Code tool

All three produce the same callable workflow - which becomes a tool the moment you publish it.

●

Record it

Do the task once in a real browser; the recorder captures every step into a workflow.

✦

Describe it

Tell an AI session your goal in plain language; it drives the page to build the workflow for you.

⌖

Discover it

Let API discovery watch a site's own network calls as you browse, then turn them into a fast workflow.

Get started free See workflows
also ▸ other clients 06
06
other clients

Use Writ tools in any MCP client

MCP clientClaude Desktop MCP tools ▸Give Claude Desktop real web actions on any site (login, forms, multi-step tasks) over MCP.MCP clientCursor MCP tools ▸Give the Cursor agent real web actions on any site via MCP, without leaving the editor.MCP clientWindsurf MCP tools ▸Give Cascade real web actions on any site via MCP, authenticated and on your own accounts. OverviewAll MCP tools ▸How Writ exposes any website as authenticated, action-taking MCP tools. For developersREST API ▸Every workflow is also a callable REST endpoint, not just an MCP tool.
faq ▸ questions answered 07
07
faq

Claude Code + Writ questions

How do I add Writ tools to Claude Code?
Run the claude mcp add command; -e puts your wt_ key in the environment, not in the process list or your shell history. Confirm the server is connected with claude mcp list. Your Writ workflows are now callable tools in the session. Each workflow you publish in Writ appears as a tool, selected by its description and typed input schema.
What can Claude Code actually do with a Writ MCP tool?
It takes real actions on a website: log in, fill forms, click through multi-step flows, extract fields, and submit - then return structured JSON. Because each tool is a real browser workflow, it works on sites that have no API. Writ runs on your own accounts, with your own credentials and data, on sites you are authorized to use.
Does it handle login and 2FA in Claude Code?
Yes. Writ runs the workflow in a real browser on your own account, handling login, TOTP and mailbox-OTP two-factor, and dynamic pages - so the tool completes authenticated tasks the agent could not reach otherwise.
Is this read-only, like most website-to-MCP servers?
No. Most website-to-MCP tools only read page text. Writ tools act - they log in, fill, click, and complete multi-step tasks, then return typed output. Claude Code calls the tool; Writ runs the real workflow on your account.
How is a tool authorized?
Every call carries a scoped wt_ API key. You can issue a separate key per project or per Claude Code setup, and revoke it without touching the rest.
rec ▸ ready ●

Give Claude Code tools that act

Publish a workflow once, then call it from your agent as an authenticated, action-taking MCP tool. Get started free.

Get started