MCP Setup

Valley speaks MCP (Model Context Protocol), the open standard AI assistants use to connect to outside tools. Once connected, you can ask your assistant about your campaigns, prospects and replies in plain language, and let it act on your behalf.

Things people ask once they are set up:

• "Which campaigns are running, and is anything stuck?"

• "Why has nothing been sent since Tuesday?"

• "Show me replies from this week that look interested."

• "Draft a reply to Sarah, in my usual voice."

• "Pause the Q3 campaign."

Before you start

You need a Valley account and access to at least one workspace. Everything the assistant can see or do is limited to the workspaces you personally have access to, with the same permissions you have in the app. Connecting an assistant never widens what you can reach.

Get your connection address

1. Open Valley and go to Settings → MCP.

2. Copy the address shown at the top of the page.

That page always shows the correct address for your account, along with ready-made snippets for each tool. It is the source of truth: prefer copying from there over typing an address from memory.

Connect your assistant

Claude

1. Copy your Valley address from Settings → MCP.

2. In Claude, open Settings → Connectors → Add custom connector.

3. Paste the address and click Add. A browser window opens: sign in to Valley and choose the workspace you want to connect.

4. In any chat, click + → Connectors, and switch Valley on.

No file editing and no keys to copy around. Claude connects with the address and a browser sign-in.

ChatGPT

1. Copy your Valley address from Settings → MCP.

2. In ChatGPT, open Settings → Connectors, and turn on Developer mode under Advanced.

3. Click Create, paste the address, and save. Sign in to Valley when the browser window opens.

4. In a new chat, open the tools menu and turn Valley on.

No Create button? On business plans, a ChatGPT workspace admin has to enable connectors first.

Cursor

1. Copy the ready-made snippet from Settings → MCP. No key needed.

2. In Cursor, open Settings → MCP → Add new MCP server.

3. Paste the snippet into the file that opens, and save.

4. Click the login prompt next to Valley, sign in, and choose your workspace. The tools appear automatically.

Any other MCP client

Valley exposes a standard Streamable HTTP MCP endpoint and supports two ways to authenticate:

• Browser sign-in (recommended). Point your client at the address. Clients that support remote MCP sign-in will open a browser, let you pick a workspace, and handle the rest. Nothing secret is stored in your config file.

• Workspace key. For tools without browser sign-in, generate a workspace key in Settings → MCP and send it as an x-api-key header.

Treat a workspace key like a password. Anyone holding it can act in that workspace as you. Generate one only if your tool cannot do the browser sign-in, and replace it immediately if it is ever exposed.

Choosing read-only or full access

When you sign in, you choose what the connection is allowed to do:

• Read-only. The assistant can look at campaigns, prospects, replies and analytics, and can draft messages for you to review. It cannot change anything.

• Full access. The assistant can also make changes, such as pausing a campaign, approving a message, or sending a reply.

Read-only is the safer starting point, and you can create a second, fuller connection later. A read-only connection can never be widened after the fact by the assistant: to change it, disconnect and connect again.

Working across several workspaces

If you belong to more than one workspace, one connection can cover several of them. Choose the workspaces at sign-in, and afterwards ask your assistant to work in a specific one by name.

Access is always checked live. If you leave a workspace, or your role changes, the connection reflects that immediately, without needing to reconnect.

What your assistant can do

Once connected, the assistant sees a catalogue of Valley tools. Broadly:

The exact list you see depends on your role and on what your workspace admin has enabled. Settings → MCP shows the live catalogue for your account.

Safety behaviour worth knowing

• Actions that matter ask first. Anything destructive, anything that sends a message to a real person, and anything that spends credits requires an explicit confirmation before it runs. Your assistant will show you what it is about to do and wait. If your client cannot display a confirmation prompt, those tools are hidden rather than run unconfirmed.

• Everything is recorded. Every change made through a connection is written to your workspace's activity trail, including which connection made it. You can review it in Settings → MCP.

• Admins are in control. A workspace admin can turn off changes entirely, so connections become read-only for everyone, or disable individual tools. If a tool you expected is missing, this is the usual reason, and the assistant will say so when it is.

• Your assistant reads your data, and data can be untrustworthy. Prospect profiles and inbound replies are written by other people. Valley marks that content as untrusted so your assistant treats it as information rather than as instructions, but it is worth reviewing anything consequential before approving.

Disconnecting

Open Settings → MCP and revoke the connection. It stops working immediately. Revoke and reconnect any time you suspect a device or config file has been exposed; if you used a workspace key, generate a new one as well.

Troubleshooting

The assistant cannot see Valley at all. Confirm the connector is switched on for that specific chat. Claude, ChatGPT and Cursor all require enabling it per conversation as well as adding it once in settings.

"Tool not found" for something you expected. Usually either the connection is read-only, or an admin has disabled that tool. Check Settings → MCP.

It was working and stopped. Sign-ins expire, and connections end when your access to a workspace changes. Reconnect from your client and sign in again.

A campaign will not send. Ask the assistant to diagnose the campaign. It checks the whole chain, including LinkedIn connection health, prospects, and the approval backlog, and reports what is blocking.

Still stuck. Contact support from the app with the name of the tool you were using and roughly when it happened, and we can trace it in the activity trail.