Getting started
Create an account, connect your coding agent, and author your first Feature
Specbench runs in your browser — there's nothing to install. Setup comes down to three things: create an account, create a Project, and (optionally) connect your coding agent.
Create an account
Sign up
Head to the Specbench web app and create an account. You can sign in with email or your existing identity provider.
Create a Project
A Project is the container for all your spec content — every Workstream, Task, and artefact you author lives inside it. Create one from the projects page; the in-app Setup Checklist will walk you through the rest.
Invite your team
Specs are better authored together. Invite teammates from the organisation's users page — product managers and domain experts are first-class authors here, not just reviewers.
Connect your coding agent
Specbench exposes its full authoring surface through a hosted MCP server over streamable HTTP with OAuth. Any MCP-capable agent can connect — authentication runs through the same account you use in the browser.
For Claude Code:
claude mcp add --transport http specbench https://mcp.specbench.com/mcpThen run /mcp inside Claude Code to complete the OAuth sign-in. For other
clients (Cursor, Claude.ai connectors, ChatGPT), add
https://mcp.specbench.com/mcp as a remote MCP server and follow the
sign-in prompt.
Once connected, your agent can read the spec, draft artefacts, raise decision points, and trace the graph — against the same project your team sees in the browser.
Your first spec
Open a Workstream
A Workstream is where authoring happens. Think of it as a long-lived thread for a slice of product work: drafts, conversation, and iteration accumulate here without touching your repository.
Author a Feature
Inside the Workstream, create a Feature and add Scenarios. Each Scenario is a sequence of Given/When/Then steps:
Scenario: Customer applies a discount code
Given a customer with an active cart
When they apply the code "WELCOME10"
Then the cart total is reduced by 10%You can edit steps individually, or paste a whole Gherkin block and Specbench parses it into structured steps for you.
Model the domain
Go as deep as your work needs: define Bounded Contexts, Aggregates, Value Objects, Use Cases, Events, and the rest of the artefact catalogue. Everything is linked, and the graph view lets you trace flows across the model.
Ship it with a Task
When a slice of your drafts is ready to deliver, bundle it into a Task. Moving
the Task from Draft to Ready merges the selected spec changes to your
configured target — the Task is the contract your implementation is built
against.
Sync to your repository
Connect a GitHub repository from the organisation's integrations page and use the manual Sync action to write your shipped specs into it. Detailed sync setup instructions coming soon.
What about running locally?
Not yet — Specbench is currently hosted-only. A local CLI, a desktop app, and an open-source self-hosting story are all planned, but for early access we're focused on the browser and the MCP server.
Need help?
Join the Specbench Discord and ask in the support channel — while we're in early access, that's the fastest way to reach us.