How many tools does it actually expose?

Ten today. The package registers more, but the surface is revealed in waves and the committed default is the core wave plus the six always-on auth tools. You can opt into more with INTEMPT_MCP_TOOLS, and the enabled set is read-only until the write wave ships.

Do I need an API key?

No. Run the login tool in the chat and a browser opens on your normal Intempt sign-in, using authorization code with PKCE over a loopback redirect. The token refreshes itself.

What is the difference between the MCP server and the CLI?

The MCP server reads your live platform data from inside an agent conversation. The CLI is a build-time tool that scans your codebase and generates typed SDK wrappers. Different jobs.

MCP server · v1.2.0

Read your customer datafrom inside Claude Code.

Your segments, users and attributes, readable from inside the agent you already use. One command to add, browser login, no API key.

Add the server Read the docs

Free to install. Works with any MCP client you already have open.

A brass deep-sea plug about to click into a porthole socket.

Segment

Trial users, 14-day inactive

Live user count

1,284

which trial users went quiet this month

•Acme Corp

•Northwind Traders

•Globex Labs

•Initech

•Umbra Studios

list_segment_users

$ claude mcp add intempt -- npx -y @intempt-technologies/mcp

Same server,every agent you already run.

Claude Code, Claude Desktop, Cursor, Windsurf, VS Code with Copilot, Zed and Cline. Stdio and HTTP transport, so desktop and remote clients both work.

Claude Code

# mcp connected

$ list_segments

→ 14 segments

Claude Desktop

list all trial users

list_users→ 42

Cursor

// mcp: intempt

await intempt.list_segments()

// → Segment[14]

Windsurf

// mcp: intempt

await intempt.list_segments()

// → Segment[14]

VS Code

// mcp: intempt

await intempt.list_segments()

// → Segment[14]

Zed

// mcp: intempt

await intempt.list_segments()

// → Segment[14]

Cline

// mcp: intempt

await intempt.list_segments()

// → Segment[14]

Describe the outcome.It picks the right tool.

Nine skills route the request instead of you picking an endpoint. Zero of them guess. All of them cite what they called.

which accounts have power users who went quiet this month?

Ask

skill selected → /analyze

tool call → list_segment_users(segment_id: power_users, activity: <30d, silent: true)

result ← 12 accounts returned

Ten tools live today,nothing behind a curtain.

The package registers 125 tools. The default core wave serves 10. Everything else is gated behind waves you opt into with INTEMPT_MCP_TOOLS.

The enabled surface is read-only. It can list segments, users, segment members and attribute columns. It cannot create a journey. That is still console work, and the page tells you so.

Nothing enters a wave on the strength of its unit tests alone.

Tools served

Registered in code

125

Domains covered

segments · users · journeys · analytics · meetings · deals · brand · registry

Operations in largest domain

journeys, including experiments and drafts

Every question your team asksin the console, in a chat.

Eight prompts, eight real domains, one command to install.

accounts · users

“Which accounts have power users who went quiet this month?”

Resolves against the real accounts and users domains, no export.

segments

“Create a segment of trial users who haven't converted in 14 days.”

One call, live, ready to route into a journey.

analytics

“Get the funnel drop-off for our signup flow this quarter.”

The same data your dashboards already show.

meetings

“Summarize my last three meetings with Acme Corp.”

Meeting summaries and activity, not a guess.

outreach

“Draft outreach for accounts that just hit power-user status.”

The same drafter Blu runs in-app.

journeys

“List every journey with a live A/B experiment right now.”

Journeys is the largest domain, 43 operations.

brand

“What's our brand voice guidance for outbound emails?”

Real brand-kit and persona data.

deals

“Pull deal activity for every open deal over $50k.”

No CRM export required.

Two ways to add it.Both take one command.

Option A

claude mcp add, done.

One command in your terminal. Node 18+, no dotfile, no marketplace step. All nine skills self-register into ~/.claude/skills via a postinstall hook.

claude mcp add intempt -- npx -y @intempt-technologies/mcp

Claude CodeClaude DesktopCursorWindsurfVS CodeZedCline

Option B

The Claude plugin marketplace.

Rather manage it through the plugin marketplace? Same server, same skills, one command from inside the chat.

/plugin install intempt@intempt-plugins

Opt out of skill install with INTEMPT_SKIP_SKILL_INSTALL=1.

Wired into the toolsyour team already opens.

Salesforce, HubSpot, Slack, Gmail, Calendar, Zoom, Segment, Shopify. Blu Agent operates them for you without a browser tab.

See the full list

Teams love it

They stopped stitching tools andhired one system.

Acquisition and retention are one connected motion. Personalizing the path before signup only pays off when the same system carries the user through onboarding afterward.

Anonymous traffic was not turning into registered users, and acquisition and retention were run as two separate problems. We ran Experiences on the anonymous visitor flow to put the right content and CTAs in front of each visitor, then handed new signups to automated Journeys for onboarding and deeper engagement.

455,396

users instrumented

Read the full research

Judge experiments on revenue per visitor instead of conversion rate. A variant can win on conversions and still lose on revenue, and repeat buying only shows up when the whole journey is in one view.

High-intent shoppers browsed a large, specialized outdoor catalog, and browsing, product interest and purchase history sat in different places. We unified them into one customer view, then ran real-time personalization, intent-based segmentation and timely follow-ups on top of it, judging experiments on revenue per visitor.

Read the full research

Build the pipeline on observed signals. When follow-ups trigger on intent, the team spends its time on the users who are genuinely engaging.

Static lead lists and timeline-based follow-ups were hiding who was actually engaging. We ran a signal-led outbound pipeline where follow-ups fire on real behavior: outbound replies, onboarding progress and product usage.

Read the full research

Free to install

Add the server.Ask it something.

Works with whichever MCP client you already have open. Logs in through the browser you are already signed into.

Add the server Read the docs

v1.2.0 · Node 18+ · stdio and HTTP transport.

Add the server. Ask it something.

Works with whichever MCP client you already have open. Free to install, browser login, no API key.

Nothing ships until you approve it.