Claude Code with Artifact

The agent you already trust, with a memory that isn't a chat and a reviewer that reads the whole product. Artifact doesn't replace Claude Code. It briefs it before every task and checks what it did after.

The product harness →

Before the run. After the run. Nothing in between changes.

Before: the brief

Claude Code calls Artifact over MCP and pulls the context bundle for the task: the feature, its rules, its acceptance criteria, the decisions behind it, and the boundary of what must hold. You stop pasting context into the prompt.

During: your rules

Your standards, security rules, and taste are compiled from one rule set into CLAUDE.md for a Claude Code session. Change a rule once; every repo gets it. Claude Code works a scoped change, not a free roam.

After: the check

What Claude Code shipped is checked against everything the product is supposed to be, not just the diff. Mismatches are shown with the intent they violate. You decide: regression, or intent that changed.

Add Artifact. Keep working.

  1. 01

    Add Artifact to Claude Code.

    Add Artifact as an MCP server in Claude Code: https://app.artifactstudio.ai/mcp. That's it — your agent sees Artifact's tools beside its own.

  2. 02

    Artifact serves the operations.

    Orient, context, search, work packet, verify: one operation each, served identically over MCP and HTTP. Claude Code sees them beside its own tools and calls them when a task needs them.

  3. 03

    Your rules file stops drifting.

    CLAUDE.md is compiled from your harness pack instead of hand-maintained per repo. Project overrides where a team needs them. Zero footprint on the repo itself.

  4. 04

    Work as you do today.

    Same prompts, same pull requests, same CI. The brief arrives before the task; the check arrives after. That's the whole change.

Works today: a stock Claude Code install registers the endpoint and runs a real task end to end.

Artifact MCP
artifact ›

The operations Claude Code will call, live over a sample project. Try work_packet FEA-0001.

Your tool. Your subscription. Your repo.

Your subscription

All model inference runs on the Anthropic account you already pay for. Artifact's server does deterministic work only. No metered tokens you can't predict.

Your repo

Artifact never writes to it and never keeps a copy. Claude Code reads in your session; the server keeps an index and a citation, not a codebase.

Your memory, outside the tool

The record lives in Artifact, not in Claude Code's session. Switch to another host tomorrow and nothing you decided is lost. Export everything, anytime, as plain files.

Things people ask about Claude Code and Artifact.

Does Artifact replace Claude Code?

No. It serves it. Claude Code writes the code; Artifact makes sure it starts from what you decided and checks what it did.

Do I have to change how I prompt?

No. The brief arrives as context Claude Code pulls itself. You'll find you paste a lot less.

What if Claude Code touches something outside the packet?

The check compares the result against the whole product, not just the target feature. A change that contradicts a rule elsewhere shows up as a mismatch, with the rule attached.

Can my human developers use the same brief?

Yes. A work packet is the same thing whether an agent or a person builds from it. Same intent, same rules, same done-criteria.

Get it in writing.

Explore a different way to make software.
We're inviting early adopters into the conversation.