For vibe coders and AI builders

You shipped it with AI.Now find who buys it.

Connect Outlevo to Claude, Cursor, Lovable, Bolt, Replit or v0. Ask in plain English for the accounts that fit, the person to talk to and a first message with its sources. Your assistant proposes; nothing is sent until you approve it.

Connect in two minutes Get a free test key →

Free sandbox · 100 fake companies · No credits used · Not affiliated with the tools named

claude + outlevo mcpsandbox key

> Find the best-fit accounts in my list and get the top one ready for a first email
● list_companies { limit: 50 }   → 37 not yet qualified
● qualify_company { company: "Northwind Analytics" }
  confirmation_required · 1 credit
> yes
● qualify_company { confirm: true }   → fit 86 · intent 52
● discover_contacts { confirm: true }   → Jordan Ellis, VP Finance ✓ verified
● propose_revenue_action { type: "outbound_step" }
  waiting for approval in Outlevo · policy v4 · 2 sources
  nothing is sent until a person approves it

Three steps

Connect, ask, approve

1
Connect
Create a test key in Outlevo and add the MCP server, or paste one prompt into your app builder.
2
Ask
Which accounts fit, who decides, what to say. Metered work quotes its credit cost before it runs.
3
Approve
Proposed emails, audiences and handoffs wait in Outlevo with their sources until you or your policy approves them.

Setup

Pick the tool you build with

Each one is documented on the developer hub, with a sandbox key so you can try it on fake data first.

Create a test key in Outlevo: Settings → API keys
Keep it in your shell as OUTLEVO_KEY
Add the MCP server, then ask in plain English

Terminal

claude mcp add --transport http outlevo https://api.outlevo.com/mcp --header "Authorization: Bearer $OUTLEVO_KEY"

Test keys work on a sandbox of 100 fake companies: no credits, nothing sent. Full Claude Code quickstart →

Create a test key in Outlevo: Settings → API keys
Put OUTLEVO_KEY in .env (gitignored)
Add the server to .cursor/mcp.json

.cursor/mcp.json

{ "mcpServers": { "outlevo": { "url": "https://api.outlevo.com/mcp", "headers": { "Authorization": "Bearer ${env:OUTLEVO_KEY}" } } } }

Test keys work on a sandbox of 100 fake companies: no credits, nothing sent. Full Cursor quickstart →

Create a test key in Outlevo: Settings → API keys
Add OUTLEVO_KEY under Supabase → Edge Function secrets
Paste the prompt below into Lovable

Prompt

Read https://developers.outlevo.com/llms.txt. Using @outlevo/sdk on the server only, list my Outlevo sandbox companies, qualify one unresearched company (quote first, then confirm), wait for the job, and show the result. Never put OUTLEVO_KEY in frontend code.

Test keys work on a sandbox of 100 fake companies: no credits, nothing sent. Full Lovable quickstart →

Create a test key in Outlevo: Settings → API keys
Put OUTLEVO_KEY in the server environment, without a VITE_ or NEXT_PUBLIC_ prefix
Paste the prompt below into Bolt

Prompt

Read https://developers.outlevo.com/llms.txt. Using @outlevo/sdk on the server only, list my Outlevo sandbox companies, qualify one unresearched company (quote first, then confirm), wait for the job, and show the result. Never put OUTLEVO_KEY in frontend code.

Test keys work on a sandbox of 100 fake companies: no credits, nothing sent. Full Bolt quickstart →

Create a test key in Outlevo: Settings → API keys
Add OUTLEVO_KEY in Tools → Secrets
Paste the prompt below into the Replit agent

Prompt

Read https://developers.outlevo.com/llms.txt. Using @outlevo/sdk on the server only, list my Outlevo sandbox companies, qualify one unresearched company (quote first, then confirm), wait for the job, and show the result. Never put OUTLEVO_KEY in frontend code.

Test keys work on a sandbox of 100 fake companies: no credits, nothing sent. Full Replit quickstart →

Create a test key in Outlevo: Settings → API keys
Add OUTLEVO_KEY in the Vercel project's environment variables (server only)
Paste the prompt below into v0

Prompt

Read https://developers.outlevo.com/llms.txt. Using @outlevo/sdk on the server only, list my Outlevo sandbox companies, qualify one unresearched company (quote first, then confirm), wait for the job, and show the result. Never put OUTLEVO_KEY in frontend code.

Test keys work on a sandbox of 100 fake companies: no credits, nothing sent. Full v0 quickstart →

Things to ask

Prompts that work on day one

>Which of my target accounts fit the ICP best, and why? Quote the pages.
>Qualify these ten domains. Tell me the credit cost first.
>Find the finance decision maker at Northwind and verify the email.
>Draft a first email to the top account and send it for approval.
>How many credits would researching 50 accounts cost?
>Add a Readiness Card to my app's account page.

What your assistant can call

Twelve MCP tools

get_context · get_icpYour approved ICP and business facts

list_companies · get_companyAccounts with their evidence and status

qualify_companyFit against your ICP, quoted first

research_companyStandard or deep research

discover_contacts · verify_contactDecision makers, with verified email

estimate_creditsWhat an operation would cost, free

propose_revenue_actionA proposal only: a person or your policy decides

list_revenue_actions · get_jobWhat is waiting, running or done

The part that is different

Your assistant can propose. It cannot send.

Autonomous agents send because nobody stops them. Through Outlevo, every outbound step, ad audience and CRM write is checked against a policy you approved, and anything new waits for a person.

No API call or MCP tool can approve, launch or send
Live keys are refused in the browser, so a leaked frontend cannot spend your credits
Every action is recorded with the policy and sources behind it

Waiting for your approval3 of 6

ToJordan Ellis · VP Finance, Northwind Analytics SubjectMonth-end close with a new RevOps team

Jordan, congratulations on the RevOps Manager opening¹. Teams at your size often find month-end close is where a new RevOps hire spends their first quarter². Worth 20 minutes to compare notes?

1 northwind.example/careers · Oct 14 2 northwind.example/customers · Oct 12

Policy check4 of 4 passed

United States, an allowed country
Contact verified, not suppressed
12 of 40 sends today on this mailbox
Every claim has a source

ApproveEditHold

drafted 09:41 · policy v4 · 2 sources · fictional example

Questions builders ask

Do I need to write code?

No. In Claude Code or Cursor you add the MCP server and ask in plain English. In Lovable, Bolt, Replit or v0 you paste one prompt and the builder writes the integration for you.

Can my assistant send emails on its own?

No. MCP tools and the API can read, qualify, research and propose. Approving, launching and sending only happen in Outlevo, by a person or by a policy your team approved.

What does it cost to try?

Nothing. A test key works on a sandbox of 100 fake companies and five Readiness Cards, uses no credits and sends nothing. Paid work quotes its credit cost before it runs.

Where does the API key go?

On the server only: an Edge Function secret in Lovable, Secrets in Replit, a server environment variable elsewhere. Live keys are refused when sent from a browser.

Which assistants work with the MCP server?

Any client that supports remote MCP servers over Streamable HTTP with a bearer header. We document Claude Code and Cursor; app builders use the SDK through the prompt instead.

Try it on fake data first

A test key gives you a sandbox of 100 companies. No credits, nothing sent.

Open the sandbox Developer hub →

sent 14:02 · approved by Dana (owner) · policy v4 · 3 sources · Northwind Analytics