Connect your own AI assistant

Point Claude, Codex, or another MCP-compatible assistant at your agent, and run your prospecting from wherever you already work.

Everything your Ninja can do in the app, your own assistant can do too — read your pipeline, explain why a lead was ruled out, change your targeting, adjust your cadence, approve a message. It's the same set of tools; only the chat window is different.

Set it up

In the Ninja app, open Settings → Connections, find AI assistants (MCP) and click Set up.
Pick the app you're connecting — Claude, Codex, or Custom — and copy the server URL it shows you.
Paste that URL into your assistant (the three sections below say where).
Your assistant sends you back to the Ninja to sign in and pick the agent it should act on. Approve, and you're connected.
Ask it "what did my agent do today?" to confirm.

There is no key to copy and nothing to keep safe: the URL is the whole configuration, and you prove who you are by signing in, the same way you'd connect any other app.

Claude

Settings → Connectors → Add custom connector. Name it The Sales Ninja, paste the URL, and click Add. Claude will walk you through signing in. Web, desktop and mobile all share it — the connector follows your account.

Codex

Connect to a custom MCP, set the type to Streamable HTTP, and paste the URL. Leave the bearer-token and header fields empty — signing in replaces them.

Any other assistant

Claude Code, Cursor, Gemini CLI and anything else that speaks MCP take the same URL. If your client supports signing in to an MCP server, this works; if it only supports a fixed API key, ask us and we'll issue one.

What it can do

The same things you'd ask your Ninja for:

Read — today's activity, your pipeline, a lead's history, why someone wasn't qualified, how a flow is performing, how much of your daily allowance is left.
Write — change your ICP, add or edit learned rules, adjust the cadence, set your daily volume, add a lead by profile URL, pause or resume a flow, regenerate a draft, approve a message.

What you're agreeing to

The consent screen is worth reading rather than clicking through. A connected assistant can do everything you can do in the Ninja chat — including approving a prepared message, which sends it from your LinkedIn account.

It acts as you, on the one agent you picked. It can't reach a teammate's agent, another organisation, or anything else on your account.
Everything it does is recorded the same way your in-app activity is — and under your name, so a shared workspace can see who did what.
Keep your assistant's per-tool approval prompts on. That prompt is what stands between "draft something" and "send something": in the app your Ninja asks you before it writes, while a third-party assistant only asks if its own settings say so.

Disconnecting

Settings → Connections → AI assistants (MCP) lists every assistant currently connected, and when each was last used. Disconnect one and it stops working on its very next call — do that if a laptop goes missing, if you stop using an assistant, or if you're simply not sure.

Losing access to the workspace disconnects them too: if a teammate leaves your team, their assistants stop working without anyone having to remember.

Common questions

Do I need a particular plan?
No — it's available on both Starter and Pro.

Does this replace the Ninja in the app?
No, it's the same agent reached a different way. Use whichever is closer to hand; both see the same pipeline, and changes made in one show up in the other.

I have two agents. Can I connect both?
Connect the assistant once per agent — each connection is signed in separately and pinned to the agent you chose at that moment. Reconnecting the same assistant moves it to whichever agent you pick.

Can my assistant send messages to prospects on its own?
It can call the same approve-and-send tool you can. Whether it asks you first is a setting in your assistant, not ours — which is why the approval prompt matters, and why disconnecting is one click.

Someone on my team connects their own assistant. Does it act as me?
No. Each person signs in as themselves, so each connection carries that person's own access and their own name in the activity log.

Autonomy and approvals — what sends automatically and what waits for you
Security and privacy — what the product accesses and stores