Weav

Weav MCP

Learn how to connect AI tools to your Weav workspace with the Model Context Protocol (MCP). Authenticate with OAuth, then search conversations, help customers, and update training data.

Model Context Protocol (MCP)

Connect AI tools to your Weav workspace so they can search conversations, help customers, and update your knowledge base.

The Weav MCP server lets AI tools access your workspace through the Model Context Protocol. After a workspace admin or owner completes OAuth consent, those tools can work with conversations, customers, and training data in that workspace.

Use it to:

  • Search and open support conversations from an AI assistant
  • Look up customer profiles and recent history
  • Reply to conversations or update status, priority, and assignee
  • Search, add, resync, or delete knowledge base items

Only a workspace admin or owner can finish consent. Other teammates can start the connection, but Weav will ask an admin or owner to approve it.


What is Model Context Protocol?

MCP is a protocol that lets AI tools connect to Weav in a secure, standardized way. It gives models a structured way to:

  • Find and retrieve Weav data (conversations, customers, and knowledge base items)
  • Call specific tools Weav exposes
  • Keep workspace context while you work in an AI assistant

How MCP works

Weav hosts a remote MCP server that follows the authenticated remote MCP specification. The server handles requests from AI tools and exposes workspace data through a secure interface.

Connection URL:

https://mcp.weav.com/mcp

When an AI tool needs access to Weav:

  1. The tool connects to https://mcp.weav.com/mcp.
  2. Weav opens a browser consent screen. Sign in if you are not already signed in.
  3. A workspace admin or owner approves access.
  4. The tool can call Weav tools for that workspace until the connection is revoked.

Use the full URL, including /mcp. Discovery metadata advertises this endpoint; the hostname alone is not the MCP server.


Benefits of using MCP

  • Secure access — All data access is authenticated. Consent is limited to admins and owners, and tokens are scoped to one workspace.
  • Standardized interface — The same connection pattern works across MCP-compatible AI tools.
  • Workspace context — Assistants can work with your inbox, customers, and knowledge base instead of copying data by hand.
  • Faster support work — Triage conversations, reply to customers, and update training data from the tools you already use.

Available tools

The Weav MCP server provides 11 tools. Search tools return up to 50 results (default 15). IDs are workspace-scoped UUIDs.

Conversations

search_conversations

Search support conversations the same way the Weav inbox does. Only conversations with at least one message are returned. Spam is excluded unless spam is true.

Parameter Required Type Description
query No string Search subject, customer name, or email. Max 255 characters.
status No string open or closed.
channel No string email or chat.
spam No boolean When true, return spam conversations instead of inbox conversations. Default false.
limit No integer Maximum results (1–50). Default 15.

get_conversation

Get a conversation, including recent messages.

Parameter Required Type Description
conversation_id Yes string Conversation UUID from search_conversations.

send_reply

Post a reply to a conversation as a workspace teammate.

Parameter Required Type Description
conversation_id Yes string Conversation UUID.
content Yes string Reply body.

update_conversation

Update a conversation’s status, priority, or assignee. This tool does not delete conversations.

Parameter Required Type Description
conversation_id Yes string Conversation UUID.
status No string open or closed.
priority No string low, normal, high, or urgent.
assignee_id No string User or agent UUID. Omit to leave unchanged. Send null to unassign.
assignee_type No string user or agent. Required when setting assignee_id.

Customers

search_customers

Search customers by name or email.

Parameter Required Type Description
query No string Search name or email.
limit No integer Maximum results (1–50). Default 15.

get_customer

Get a customer profile, company, and recent conversations.

Parameter Required Type Description
customer_id Yes string Customer UUID from search_customers.

Knowledge base

search_training_data

Search knowledge base (training data) items by title, description, type, or source.

Parameter Required Type Description
query No string Search title, description, or source.
type No string text, url, file, video, or qa.
limit No integer Maximum results (1–50). Default 15.

get_training_data

Get a single knowledge base item.

Parameter Required Type Description
training_data_id Yes string Training data UUID from search_training_data.

add_training_data

Add knowledge base content as text, Q&A, or a website URL. Every agent in the workspace is given access automatically.

MCP can add text, qa, and url sources. File and video sources can be searched if they already exist in Training; they cannot be uploaded through this tool.

Parameter Required Type Description
type Yes string text, qa, or url.
content When type is text string The text to train on.
title No string Optional title. Max 255 characters.
question When type is qa string The question.
answer When type is qa string The answer.
url When crawling a site string Website URL to crawl. Weav discovers pages on the site and trains on them. Max 2048 characters.
urls When training specific pages string[] Specific page URLs to train without crawling the rest of the site. Max 100 URLs.

Use url for a whole website. Use urls when you only want listed pages.

delete_training_data

Delete a knowledge base item. For website/URL sources, this also deletes discovered URLs and chunks.

Deleting training data cannot be undone from the MCP client.

Parameter Required Type Description
training_data_id Yes string Training data UUID.

resync_training_data

Re-scrape an existing URL knowledge base item.

Parameter Required Type Description
training_data_id Yes string UUID of a URL training data item.

Setting things up

Authentication

The MCP server uses OAuth. When you connect, Weav opens a browser consent screen.

  • Only a workspace admin or owner can finish consent
  • Access is scoped to that workspace
  • Admins can disconnect a client later in Settings

There is no API-key or bearer-token shortcut. Connect through OAuth.

Path: Settings → Developers → MCP connections

  1. Open Settings.
  2. Go to Developers.
  3. Select MCP connections.
  4. Review connected clients, or disconnect one to revoke access immediately.

Configuration examples

The example below is a generic template. Always refer to your AI tool’s official documentation for the latest MCP setup steps. Details vary between tools and versions.

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

After you save the config, restart the AI tool. The first connection opens a browser window for Weav sign-in and consent.


Permissions

The Weav MCP server issues tokens with the mcp:use ability. That ability is granted when an admin or owner completes consent for the workspace.

If consent is declined, or the connection is disconnected in Settings, tool calls fail until someone reconnects.


Debugging and troubleshooting

Authentication problems

If the browser consent flow is stuck or the client keeps asking you to sign in:

# Stop existing mcp-remote processes
pkill -f mcp-remote

# Clear the local MCP auth cache
rm -rf ~/.mcp-auth

Then restart the AI tool and connect again.

Connection testing

# Starts the OAuth browser flow
npx mcp-remote https://mcp.weav.com/mcp

View active MCP connections

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

Error handling

  • Authentication failures — Restart the OAuth flow. Confirm a workspace admin or owner completed consent.
  • 401 Unauthorized — The access token is missing, expired, or the connection was revoked in Settings.
  • Tool errors — Confirm the IDs you pass belong to the connected workspace.

Troubleshooting tips

Check that:

  • The URL is https://mcp.weav.com/mcp (including /mcp)
  • You restarted the AI tool after changing the config
  • A workspace admin or owner completed the browser consent screen
  • The client still appears under Settings → Developers → MCP connections
  • Conversation, customer, and training data IDs are UUIDs from this workspace

If a client is stuck, disconnect it in Settings and connect again.