Skip to content
Weavz
Menu

MCP Workspaces

Add your apps, choose private or shared accounts, and connect an AI client through one MCP endpoint.

An MCP Workspace brings your apps, connected accounts and allowed actions together behind one MCP endpoint. Create it once, add the apps you need, then connect Claude, ChatGPT, Codex or another MCP client. There is no separate server to create.

MCP Workspaces and connected accounts are unlimited on every plan. Execution, storage, team membership and other plan allowances still apply. See pricing.

Connect your own AI

  1. Choose For your AI during signup. Weavz prepares your first MCP Workspace; you do not need to create a Project or workspace manually.
  2. Open the full Apps catalog, search for an app, and open its setup page. Connect the account your AI should use.
  3. Open Home or AI clients and select your AI client. Follow Guided setup and the illustrated walkthrough, or copy the setup prompt for a supported coding agent. Complete Weavz sign-in and approve the requested access.
  4. In your AI client, ask it to list the available apps or perform one small task. Opening an install link or copying a URL alone does not confirm a working connection.

The URL shown for a selected MCP Workspace always connects to that workspace. Codex, Cursor and OpenCode 1.x configuration uses a distinct connection name for each workspace endpoint, so adding another workspace does not replace the earlier entry. The generic hosted connector URL, https://platform.weavz.io/mcp/weavz, asks you to choose a workspace during authorization when you have several. Reconnect or authorize another connection to use a different workspace; an existing token does not switch workspaces.

Choose whose account an app uses

For you: each person connects a private account. Their AI uses that person's account and permissions. Another teammate's credentials are not used as a fallback.

For your team: an owner or admin selects and connects one account. Every authorized caller uses that same account and its permissions. Share an account only when those callers should be able to act through it.

Under Advanced, a shared fallback can use a person's account when connected and a selected shared account otherwise. That fallback gives callers access to the shared account when they do not have their own connection. Keep it off unless that is the intended behavior.

These choices control credentials. File and state sharing are separate app settings and are not changed by selecting a credential option. Approval rules and saved inputs are also optional advanced features.

Projects and product setup

A Project groups your team members, billing and MCP Workspaces. In For your product, use the Project menu to create or switch Projects. Weavz prepares the underlying execution workspace for your app and customer setup; detailed MCP Workspace controls live under Advanced. Each Project keeps its own members and plan.

In For your AI, Project controls are under Team & billing. Your app and AI-client setup stays in the main navigation. You can switch experiences from the menu at the top of the sidebar without moving accounts or changing access.

Existing Project identifiers and API/SDK contracts remain compatible with earlier versions.

Build a product

Your backend can provision the same MCP Workspace through the API or SDK. Keep the Weavz API key on the backend.

typescript
import { WeavzClient } from '@weavz-io/sdk'
 
const client = new WeavzClient({ apiKey: 'wvz_your_api_key' })
const { workspace } = await client.mcpWorkspaces.create({ name: 'Support assistant' })
 
await client.mcpWorkspaces.addIntegration(workspace.id, {
  integrationName: 'gmail',
  alias: 'customer_gmail',
  connectionStrategy: 'per_user',
})
 
const endpoint = workspace.mcpServer?.mcpEndpoint
python
from weavz_sdk import WeavzClient
 
client = WeavzClient(api_key="wvz_your_api_key")
workspace = client.mcp_workspaces.create(name="Support assistant")["workspace"]
client.mcp_workspaces.add_integration(
    workspace["id"], integration_name="gmail", integration_alias="customer_gmail",
    connection_strategy="per_user",
)
endpoint = workspace["mcpServer"]["mcpEndpoint"]

Create end-user identities and connect their app accounts using hosted connect. OAuth clients authorize as the actual end user. For a programmatic harness, issue a separate workspace access token using the workspace ID and your application’s external user ID. No separate server ID or authentication change is needed. Several end users can share the endpoint without sharing private accounts.

typescript
const access = await client.mcpWorkspaces.createAccessToken(workspace.id, {
  endUserId: 'customer_123',
  expiresIn: 3600,
})
// Provision access.mcpEndpoint and access.accessToken privately to this user's agent.
python
access = client.mcp_workspaces.create_access_token(
    workspace["id"], end_user_id="customer_123", expires_in=3600,
)
# Provision access["mcpEndpoint"] and access["accessToken"] privately to this user's agent.

Send the access token as Authorization: Bearer to the returned MCP endpoint. Tokens do not grant REST API access. The end user must already exist in the workspace, and connected apps still use that user's private account or an explicitly shared team account.

The API paths remain /api/v1/workspaces. Existing client.workspaces SDK methods remain supported aliases. The response includes workspace.mcpServer with its ID, mode, authentication mode and URL. Use configureMcp in TypeScript or configure_mcp in Python to intentionally change endpoint settings.

Advanced endpoint settings

Open Advanced → MCP Workspaces, select the workspace, and open its Connect your AI tab and use How your AI uses tools to select Code Mode or Tool Mode, then choose Save tool mode. The same selector is available in Endpoint settings. Project owners and admins can save this setting.

Code Mode lets an AI discover apps and combine calls in a script. Tool Mode exposes each enabled action as an individual tool. The MCP URL stays the same. Refresh tools in your AI client after switching; moving to Code Mode may require reconnecting to authorize code access. A client-specific public connector keeps its supported tool interface.

Authentication, tokens and tool restrictions are under Advanced access and compatibility → Endpoint settings. New MCP Workspaces default to Code Mode and OAuth. Project members and explicitly pre-added end users can authorize; private credentials remain bound to the caller.

Use a separate MCP Workspace when you need a separate tool surface or account policy. Changing the primary endpoint's mode or permissions affects clients using that endpoint.

Existing workspaces and clients

Existing workspace IDs, endpoint URLs, tokens, app accounts, tool restrictions, approval rules, saved inputs and Skills remain in place. If a workspace previously had several endpoints, one becomes primary. The other configurations remain under Advanced → Legacy connections, retaining their existing URLs and permissions.

Deleting the primary endpoint alone is blocked. Delete the MCP Workspace explicitly when you intend to remove it and its data. If an older endpoint has a default end-user identity, explicitly change that binding before deleting the end user.

For advanced API compatibility details, see MCP endpoint management.