AI writes the code. Spur validates it.

Spur MCP is the glue in your AI SDLC, it wires Cursor, Claude, and Copilot to your validation agent. Tests generated from every PR, regressions run from chat, bugs fixed from a link.

From a PR description to a passing test.

Point Claude Code at a pull request. It reads the diff, walks your codebase, drafts plain-English test steps, and saves them to Spur with create_test all before you've merged.

See the workflow guide

Run regressions from chat.

One sentence triggers list_tests and run_tests. The agent picks the env, reports back with timings.

Logs, network, screenshots.

Every artifact the agent needs to know if it's a test issue or a real bug, one tool call away.

Context from where work lives.

Pair Spur with GitHub, Jira, Linear or Slack. The agent matches changes to the tests that cover them.

Tests ship with your code.

Forget XPaths and accessibility IDs. Spur looks at the screen, the same pixels your customer sees, and figures out where to tap, what to type, and what to verify. The result: tests that work on any platform, any layout, any device.

01

A PR, a ticket, a commit.

Point Claude at whatever you're working on. GitHub, Jira, Linear any source of "what changed".

The whole codebase. Not just the diff.

Claude walks the surrounding files, the related state, and your existing coverage. It surfaces the primary flow, the edge cases, and the related areas a diff alone would miss.

Plain-English steps.

The output is human-readable. You can edit a word; you don't have to debug a selector.

Saved to Spur, with approval.

Anything that writes asks first. Read-only calls auto-approve so the loop stays fast.

Triggered from your editor. Run in Spur's cloud.

One tool call kicks off real browsers and real devices on Spur's infrastructure. Results, timings, and artifacts come back to your chat, you never leave the editor.

Red test? It routes the fix.

If it's the test, Claude calls update_test. If it's the code, it points you at the file, with the failing trace attached.

Read your own test coverage.

The MCP exposes your full Spur workspace as read-only tools suites, tests, scenarios, run history. Ask in plain English, get answers backed by what's actually in your suite.

Two minutes. One JSON file.

The Spur MCP works in any client that speaks MCP. OAuth runs on first connect, no API keys to paste, no separate dashboard to learn.

Code was the bottleneck. Now it's validation.

AI made shipping code easy. Spur makes shipping it safely just as easy, and the whole company sees the win.

20× Release velocity

"Spur has significantly improved Wander's testing capabilities. It has allowed us to iterate and ship so much faster with confidence."

Testimonials

Spur is our first big win company-wide in implementing the use of AI agents. When we were able to share this with our greater team, everybody was almost in awe of what we were able to achieve.

FAQ

What do I need to connect Spur MCP?

A Spur workspace and a client that speaks MCP, Cursor, Claude Code, Claude Desktop, ChatGPT, or VS Code with Copilot. Add one block to your client's config file, restart, and OAuth opens on first connect. No API keys to paste, no separate dashboard.

Do I need an existing Spur test suite to get value from this?

No. If you have tests, the MCP exposes them immediately. If you're starting from zero, point your agent at a PR and it'll draft your first tests from the diff.

Which clients does it work with?

Any MCP-compatible client. We publish ready-to-paste configs for Cursor, Claude Code, Claude Desktop, ChatGPT, and VS Code Copilot.

How long before it's actually useful?

Setup is a two-minute config edit. After that it depends on your suite: reading existing coverage works immediately, and generating your first test from a PR takes about as long as the agent needs to read the diff.