Introducing Weav MCP Server: Connect Your AI Tools to Your Support Workspace

TL;DR:

• Weav now has an MCP server at mcp.weav.com/mcp, so external AI tools can read and act on your workspace directly.

• 11 tools cover conversations, customers, and your knowledge base, so an AI assistant can search a ticket, reply as your team, or add training data without anyone copying it by hand.

• Access is OAuth-only and requires admin or owner approval, scoped to a single workspace. No API keys, no way around the consent screen.

"Support is broken. AI hasn't fixed it, yet."

We're pumped to announce that Weav is taking the next step to fix support with the release of our MCP (Model Context Protocol) server. This step opens the door for our customers to manage their support in a new way that has not been done before.

What does this mean for our customers?

You can now use Weav directly from the AI tools that you are already using like Claude, ChatGPT, Grok, etc. Want to know what the most recent conversations have been? Want to train your support agent? Need to know what content is missing from your site? You can prompt your tools, or build agents to find that information for you.

All you need to do is:

• Connect the AI tools you already use with the Weav workspace. Don't worry about having to export conversations, or pasting the customer history into another tool. The agent picks up all this information for you, making analysis and actions easier than ever.

• Let an assistant pull context before it drafts anything. Want to know how many times a customer has contacted you? Or what those conversations were about? You can search a customer's conversation history and company details in one call instead of switching tabs.

• Enhance your resolution engine knowledge without opening Weav. You can add training data directly from the AI tools you're already using. Just prompt your agent to crawl a website, create a Q&A or build a plain text doc and send it to Weav without leaving your tools.

• Security is always top of mind. We've limited access to one workspace per token. Only an admin or owner can approve the connection, and admins can revoke it at any time from workspace settings. You're always in control of what data flows in-n-out of Weav.

How to connect

1. Get approval from a workspace admin or owner. Weav's OAuth flow only accepts consent from an admin or owner account. There's no API key alternative, so you can't skip this step.

2. Point your MCP client at Weav. Use mcp-remote to connect to https://mcp.weav.com/mcp.

3. Approve once in the browser. The first connection triggers a standard OAuth consent screen. After that, reconnection happens automatically.

Technical note

Weav's MCP server uses standard MCP client configuration. A typical client config looks like this:

{
  "mcpServers": {
    "weav": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.weav.com/mcp"
      ]
    }
  }
}

{
  "mcpServers": {
    "weav": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.weav.com/mcp"
      ]
    }
  }
}

{
  "mcpServers": {
    "weav": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.weav.com/mcp"
      ]
    }
  }
}

To test the connection directly:

npx mcp-remote https://mcp.weav.com/mcp

npx mcp-remote https://mcp.weav.com/mcp

npx mcp-remote https://mcp.weav.com/mcp

If you need to reset a stuck connection:

pkill -f mcp-remote
rm -rf ~/.mcp-auth

pkill -f mcp-remote
rm -rf ~/.mcp-auth

pkill -f mcp-remote
rm -rf ~/.mcp-auth

And to check what's currently connected:

ps aux | grep mcp-remote | grep -v grep

ps aux | grep mcp-remote | grep -v grep

ps aux | grep mcp-remote | grep -v grep

You have access to 11 tools, in 3 different categories:

• Conversations: search_conversations, get_conversation, send_reply, update_conversation

• Customers: search_customers, get_customer

• Knowledge base: search_training_data, get_training_data, add_training_data, delete_training_data, resync_training_data

Personally, I'm most excited for the add_training_data tool. It makes its so easy for you to add knowledge to your resolution engine. add_training_data accepts text, Q&A pairs, or a website URL to crawl, so an AI tool can expand your knowledge base the same way you'd do it manually in Weav. Tokens carry the mcp:use scope, tied to the single workspace they were approved for. For more on how Weav trains on your existing docs, see how to train an AI Agent on your product docs.

We're making this secure

Every connection requires OAuth consent from a workspace admin or owner, and every token is scoped to exactly one workspace. By design, we didn't add an API key path for this integration. If you manage multiple Weav workspaces, you'll need a separate connection and a separate approval for each one.

While a shared key is more convenient, we made this decision because it means that no token can quietly reach data outside its workspace that it was approved for. We take every measure to protect our customers data, and leaking data is a non-starter. If you need to shut off access, an admin can revoke the connection from workspace settings at any time, no waiting on us.

You can connect your first AI tool to your Weav workspace at mcp.weav.com/mcp, read the full reference at docs.weav.com/api-reference/mcp, or browse the server on GitHub.

FAQ's

What is the Weav MCP server?
It's a Model Context Protocol server that lets external AI tools securely access a Weav workspace's conversations, customer data, and knowledge base, using the same live data your team works from instead of a manual export.

Do I need to be a developer to use this?
To connect an AI tool to Weav, yes, someone needs to configure an MCP client and get OAuth approval from a workspace admin or owner. Once connected, anyone using that AI tool benefits from the connection.

Is my data safe?
Access requires OAuth approval from an admin or owner, and every token is scoped to a single workspace. There's no API key that could be copied or shared outside that approval flow.

Can I use an API key instead of OAuth?
No. OAuth consent from an admin or owner is required for every connection. This is intentional, not a current limitation.

What can an AI tool actually do once it's connected?
Search and retrieve conversations, reply as a team member, update conversation status/priority/assignment, look up customers and their history, and search, add, delete, or resync knowledge base training data. Eleven tools total, listed above.

Which AI tools work with this?
Any MCP-compatible client. Setup uses the standard mcp-remote pattern shown above.

Can I revoke access once I've approved it?
Yes. An admin can revoke an MCP connection at any time from workspace settings. Approving a connection isn't permanent.

Related articles

• Which AI Model Actually Matters for Customer Support?

• AI Model Cost: Why the Right Model Matters

• Support is broken. AI hasn't fixed it, yet.