Give your coding agent the growth stack.

One command for real growth context.

Two commands. Setup signs you in, picks your project, and connects your agent.

# install the mim command (Node 20+)
npm install -g @mimeticinc/mim-cli

# sign in and connect your agent
mim setup

Setup registers the server with whichever agents it finds: Claude Code, Codex, Cursor, Windsurf, Claude Desktop. It writes each config file, backs up what was there, and leaves any other MCP servers you have alone. Add one later with mim mcp install cursor, or re-run mim mcp install all.

Install globally, not with npx. Your agent runs mim as the MCP server, so it has to stay on your PATH. Setup prints every step and names the fix for any that fail, including the exact export PATH= line if your shell cannot find mim afterwards. Run it again whenever you like: it reuses your login rather than replacing it.

Signing in stores an expiring token in ~/.mim/config.json, with restrictive file permissions where the system supports them. It is not encrypted and not in your keychain. mim auth logout revokes it before deleting it. Prefer environment or secret-manager storage? Set MIM_API_TOKEN instead. Privacy details.

• mim projectsList your project keys. Run this first: a key looks like yourstore-com-4f2c1a, not your domain.
• mim projects use <key>Save a default project so every other command can drop --project.
• mim contextPrint a growth-context pack for a project, ready to paste into Claude Code or Codex.
• mim query-ga4Read-only GA4 report over your connected property: any metrics, dimensions, and date range.
• mim query-gscSearch Console top queries and pages: clicks, impressions, CTR, position.
• mim ads queryRead-only GAQL report on the linked Google Ads account. Defaults to 30-day campaign performance.
• mim recordingsSession-replay summaries: what each visitor did, console errors, and the likely point of friction.
• mim replay --latestOne replay's issues plus a redacted, plain-English narrative of the session.
• mim findingsThe audit's prioritized findings and backlog for a project.
• mim audit start <url> --waitRun an audit and block until the report is ready. Reuses a recent audit of the same site; audit rerun forces a fresh one.
• mim fixes listSee which findings are fixable as a pull request.
• mim fixes start <rank>Queue a code fix. Mimetic opens a PR with a preview link for you to review and merge.
• mim setup-analyticsNo analytics yet? Provisions a GA4 property, installs the tag by PR or Shopify pixel, verifies Search Console, and submits your sitemaps once verified.

There is more behind mim --help: posthog query (read-only HogQL), klaviyo query (campaign and flow performance, an abandoned-cart flow check, a dead-link check over recent campaign emails), properties to repoint a project at the right GA4 property, and ads setup.

The whole stack, in one MCP server.

This is the part most agents are missing: real data. Your agent gets project-scoped, read-only tools for your audit findings, redacted narratives of real visits, and queries against whatever you have connected: GA4, Search Console, Google Ads (GAQL), PostHog (HogQL), Mixpanel, Klaviyo, and A/B results from Intelligems, VWO, Optimizely or GrowthBook. Raw replay payloads are never returned.

# already connected by mim setup. just ask, in plain English:
> Review the last 20 recordings. Find the bugs and UX friction, and propose fixes.
> Which landing pages lost the most sessions this month, and why?

The first prompt hits a purpose-built review tool: it returns a code-review brief with per-session evidence, repeated friction patterns, and audit context, so the agent proposes changes it can defend. To ship a fix, your agent runs mim fixes start in your shell and Mimetic opens the pull request with a preview link for you to review and merge. And if a site has no analytics at all, the agent can run the hosted setup itself: a GA4 property, the tag as a PR or as a snippet the agent applies in the repo, Search Console verification, and sitemap submission.

Or call it directly.

The CLI is a thin client over HTTPS, so everything it does is available to your scripts, CI, and backends. Authenticate with the bearer token mim auth login issues, project-scoped, no scraping, no headless browser to run yourself. Core resources are plain JSON under /api/mim; analytics and provider queries go through the MCP endpoint as JSON-RPC over HTTP.

# list your projects
curl https://trymimetic.com/api/mim/projects \
  -H "Authorization: Bearer mim_xxx"

# start an audit and poll for the report
curl -X POST https://trymimetic.com/api/mim/audits \
  -H "Authorization: Bearer mim_xxx" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://yourstore.com","mode":"slim"}'

Managed tokens are scoped to projects available to the authenticated account, expire, and can be revoked with mim auth logout. Session-replay recordings require the replay add-on on an active plan.

Building an AI agent?

Mimetic exposes a pay-per-call audit over the x402 protocol, your agent pays in USDC on Base and gets a full report, no account or API key. It's even discoverable in the x402 bazaar.