Weavz MCP App
Connect Weavz directly from Claude, ChatGPT, Codex, Cursor, and other remote MCP clients.
The Weavz MCP App is the fastest way to use a Weavz workspace inside AI clients that support remote MCP connectors. Instead of creating a custom MCP server first, add the Weavz connector URL, sign in, choose the workspace the agent should use, and let Weavz expose the workspace's configured integrations through Code Mode.
Use this guide for Claude, ChatGPT, Codex, Cursor, Claude Code, and other MCP-compatible agents when the user experience should stay inside the AI client as much as possible.
Connector URL
Use your Weavz platform URL plus /mcp/weavz.
https://platform.weavz.io/mcp/weavzFor self-hosted or private deployments, replace the host with your public Weavz API host:
https://your-weavz-domain.example.com/mcp/weavzThe endpoint must be reachable by the MCP client over HTTPS. Hosted clients such as Claude and ChatGPT connect from their own cloud infrastructure, so a connector URL that only works from your laptop, VPN, or private office network will not work unless the client provides a tunnel or your network allows that traffic.
Fast Client Setup
Use the setup panel above when you are adding Weavz to a hosted AI client or a coding agent. Claude supports a prefilled custom-connector setup URL, so the Claude button opens the connector flow with the Weavz name and URL already filled. ChatGPT does not currently document an equivalent stable prefilled URL for arbitrary custom MCP apps, so open ChatGPT connector settings, paste the Weavz connector URL, and let ChatGPT start the Weavz OAuth flow.
For current client-specific behavior, review Claude's custom connector guide, OpenAI's ChatGPT connection guide, and OpenAI's developer mode guide before publishing setup instructions for your own users.
What Users See
- The user adds the Weavz connector in the MCP client.
- The client opens Weavz sign-in.
- If the user belongs to more than one workspace, Weavz asks which workspace this connector should use.
- Weavz shows the connector authorization screen.
- The client receives an MCP token scoped to that one workspace.
- In clients that support MCP app UI, the user can open the Weavz MCP App inside the client to add integrations, connect or reconnect user-owned credentials, refresh status, view pending approvals, and open the dashboard for advanced settings. Plain remote MCP clients still receive the same tools, but users manage integrations from the dashboard or connector authorization flow.
If a signed-in user has exactly one workspace, Weavz uses that workspace without showing an extra chooser. If a new user has no workspace yet, Weavz creates a starter workspace named Weavz Agent Workspace with Code Mode, Agent Browser, Agent Browser AI, Filesystem, State KV, memory, scratchpad, web reading, HTTP, transformation, date/time, hash utilities, and common SaaS integrations.
The workspace is selected during sign-in so the connector token stays scoped. To use a different workspace, disconnect or reauthorize the Weavz connector and choose the other workspace during the sign-in flow.
Choose The Right Setup
| Scenario | Recommended path |
|---|---|
| A user wants Weavz inside Claude, ChatGPT, Codex, or another remote MCP client | Use the Weavz MCP App at /mcp/weavz |
| Your team wants to publish one general Weavz connector for broad use | Use the Weavz MCP App and submit the same remote MCP endpoint where the platform supports app review |
| Your product needs one workspace-specific server with a curated URL | Create a Code Mode MCP server or Tool Mode MCP server |
| Your backend provisions MCP clients for known end users | Create an MCP server through the API and issue end-user bearer tokens when OAuth is unavailable |
The Weavz MCP App uses Code Mode by default. Agents get weavz_search, weavz_read_api, and weavz_execute for compact, dynamic use of the selected workspace. The embedded app adds a visual control surface for connection management and status.
The starter workspace exposes deterministic browser tools as weavz.browser. It also exposes Agent Browser AI as weavz.browser_ai; those natural-language browser actions require the user to connect an LLM provider key before they can run.
Common starter aliases:
| Alias | What agents use it for |
|---|---|
weavz.browser | Deterministic browser control with screenshots and human handoff |
weavz.browser_ai | Natural-language browser actions using a connected LLM provider key |
weavz.files | Filesystem artifacts and generated files |
weavz.state | Small durable key-value state |
weavz.memory | Durable agent memory |
weavz.scratchpad | Temporary notes for ongoing work |
weavz.web | Public web reading and extraction |
weavz.http | Direct API calls |
weavz.transform | Data shaping between tools |
weavz.datetime and weavz.hash | Date utilities, IDs, hashing, and encoding |
For browser work, agents should batch several browser operations inside one weavz_execute call, such as snapshot, click, type, evaluate, and screenshot, then return a concise result. Reuse sessionId across separate runs only when the workflow needs human handoff or persistent browser state.
Try It In Claude
Claude supports custom remote MCP connectors. For individual plans, use the prefilled Claude setup button above or add a custom connector from Customize > Connectors. For Team and Enterprise plans, an Owner or Primary Owner can add the connector from organization settings; members then connect it individually.
Open the prefilled setup
Click Open prefilled Claude setup in the setup panel. Claude opens the custom connector flow with the Weavz connector name and URL already filled.
Confirm the connector
If Claude opens connector settings without prefill, add a custom connector
manually and paste https://platform.weavz.io/mcp/weavz, or your self-hosted
Weavz domain plus /mcp/weavz.
Sign in to Weavz
Claude opens the Weavz authorization flow. Sign in, choose a workspace if prompted, and approve the connector.
Enable it in a chat
Enable Weavz from Claude's connector or tools menu for the conversation where the agent should use it.
Claude's remote connector docs note that hosted Claude clients connect to the remote MCP server from Anthropic's infrastructure, not from the user's local device. Review Claude's custom connector setup guide before submitting or rolling out an organization connector.
Try It In ChatGPT
ChatGPT can test remote MCP apps in developer mode, and public distribution uses OpenAI's app submission flow. For private testing, create a connector in ChatGPT settings and use the Weavz MCP App URL.
Enable developer mode
In ChatGPT, open Settings > Apps & Connectors > Advanced settings and enable developer mode if your account or organization allows it. The Create button for custom MCP apps appears after developer mode is active.
Create a connector
Open connector settings, create a new connector, and enter a user-facing name, description, and the Weavz connector URL from the setup panel.
Connect
Save the connector. When ChatGPT connects, Weavz handles sign-in, workspace selection, and token scoping.
Use it in a chat
Start a new conversation and add the Weavz connector from the composer tools menu.
For public ChatGPT distribution, review OpenAI's current Apps SDK connection guide, authentication guidance, and app submission guidelines.
Try It With Coding Agents
Coding agents that support remote HTTP MCP can use the same endpoint. Use OAuth when the client supports browser authorization.
| Client | Setup |
|---|---|
| Claude Code | Run the CLI command below |
| Codex | Add the remote server, then run browser login |
| Cursor | Add a remote HTTP MCP server named weavz |
| Generic MCP config | Use the JSON shape below |
Claude Code setup:
claude mcp add --transport http weavz https://platform.weavz.io/mcp/weavz
claude mcp listCodex CLI setup:
codex mcp add weavz --url https://platform.weavz.io/mcp/weavz
codex mcp login weavzManual Codex config:
[mcp_servers.weavz]
url = "https://platform.weavz.io/mcp/weavz"Generic MCP config shape:
{
"mcpServers": {
"weavz": {
"type": "http",
"url": "https://platform.weavz.io/mcp/weavz"
}
}
}Connect Integrations Inside The App
After the connector is authorized in a client with MCP app UI support, ask the AI client to open Weavz, or call the app's visible entrypoint from the client tools menu. The embedded app shows:
| Area | What it does |
|---|---|
| Integrations | Lists configured workspace integration aliases, connection state, and available connect actions |
| Add | Searches the Weavz app catalog and adds a new alias to the connector workspace |
| Agent Surface | Shows the MCP tool count and pending approval count |
| Connect buttons | Opens a hosted connection flow for integrations that need user-owned credentials |
| Disconnect and remove actions | Disconnects the current connector user from an alias, or removes a connector-managed alias from the workspace |
| Approval and link cards | Shows large action buttons for end-user approvals, dashboard approvals, connect links, and short-lived download URLs when the MCP client supports app UI output |
| Dashboard link | Opens advanced workspace, MCP server, approval, partial, billing, and admin settings |
Admins can configure deeper behavior in the dashboard: workspace integration aliases, connection strategies, enabled actions, input partials, Human Gates, built-in Filesystem/State KV settings, and Sandbox policy. Changes to workspace integrations sync into the connector's MCP server.
Clients that do not render MCP app UI still get the Code Mode tools. In those clients, manage integrations and approvals from the dashboard, or use connect links returned by tools when the client supports opening links.
Aliases are what agents call. Use purpose-readable names such as office_slack, support_slack_bot, customer_gmail, or billing_stripe; avoid vague names like default, prod, or slack2 when a workspace may contain more than one configured account.
Integrations added from the Weavz MCP App create a workspace integration alias, but their credential session is scoped to the signed-in connector user. This keeps one user's Slack, Gmail, Sheets, or other account separate from another user's account even when they use the same connector workspace. Built-in Filesystem, State KV, and Code Mode state created through the connector default to end-user scope as well. Use the dashboard, SDK, or API when you intentionally want a shared workspace credential or shared workspace state.
Workspace Selection
Workspace selection is intentionally part of the authorization flow.
- A connector token is scoped to one selected workspace.
- A user with more than one workspace chooses during sign-in.
- A user with one workspace skips the chooser.
- A user with no workspace receives a starter workspace.
- To change workspaces, disconnect or reauthorize the connector and choose again.
This keeps tool execution, connection resolution, approvals, and built-in state tied to the selected workspace instead of letting a running chat silently switch context.
Launch Checklist
Before inviting users or submitting the connector to a marketplace:
- Confirm the connector URL is public, HTTPS, and stable.
- Verify sign-in works for new users, existing single-workspace users, and existing multi-workspace users.
- Connect at least one OAuth integration from the embedded app.
- Confirm Code Mode tools return expected results from the selected workspace.
- Test a Human Gate approval from the MCP client.
- Confirm reconnecting lets the user choose a different workspace.
- Review tool names, descriptions, and workspace integration aliases for clarity.
- Confirm destructive or sensitive actions have Human Gates or are not enabled.
- Test in every target client before submitting for public distribution.
Troubleshooting
| Symptom | Check |
|---|---|
| The client cannot add the connector | Confirm the URL is public HTTPS and ends with /mcp/weavz |
| The user sees the wrong workspace | Disconnect or reauthorize the connector and choose the intended workspace during sign-in |
| Integrations are missing | Add workspace integrations in Weavz and refresh the connector metadata in the client |
| A connect button is disabled | The integration may use a fixed/shared connection or may not require external authentication |
| A tool call says credentials are missing | Connect the integration from the embedded app or dashboard, then retry |
| A write action pauses | Review and approve the Human Gate request, then retry the tool call |