MCP and plugins
One remote MCP server for any compatible agent, plus a Claude Code plugin with /wyatt and workspace skills. Sign in with the browser — you never paste an API key.
A Claude Code plugin, a Cursor MCP connector, and a ChatGPT app — all talking to the same Wyatt workspace.
Remote MCP
Point Claude Code, Codex, Cursor, ChatGPT, or any Streamable HTTP client at Wyatt. Approve access in the browser and the agent can search, read, and write in your workspace.
https://www.wyattworks.ai/api/mcp-server
Claude Code
The Claude Code plugin bundles the MCP server plus a /wyatt command and workspace, document, and task skills. Cursor and ChatGPT connect through MCP instead.
claude plugin install wyatt@wyatt
Pick your client. Every path ends the same way: a browser window opens, you approve access, and the agent lists Wyatt's tools.
Install the Wyatt plugin, or add the MCP server directly.
Plugin (adds the /wyatt command and skills)
claude plugin marketplace add luminpdf/wyatt-plugin claude plugin install wyatt@wyatt
The plugin bundles the MCP server plus a /wyatt command and workspace, document, and task skills.
MCP server only
claude mcp add --transport http --scope user wyatt https://www.wyattworks.ai/api/mcp-server claude mcp list
Start Claude Code and run /mcp to complete authorization.
Three commands: add, log in, verify.
Terminal
codex mcp add wyatt --url https://www.wyattworks.ai/api/mcp-server codex mcp login wyatt codex mcp list
The login step opens your browser to approve access.
One entry in mcp.json, then enable it in settings.
~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project)
{
"mcpServers": {
"wyatt": {
"url": "https://www.wyattworks.ai/api/mcp-server"
}
}
}Open Cursor Settings > Tools & MCP, enable Wyatt, and complete OAuth.
Add Wyatt as an app in developer mode.
Settings > Apps
1. Turn on developer mode under Settings > Apps > Advanced settings. 2. Choose Apps > Create and name the app "Wyatt". 3. Enter the endpoint: https://www.wyattworks.ai/api/mcp-server 4. Pick OAuth as the authentication method and approve access.
Hand your agent this prompt, or configure the endpoint yourself.
Setup prompt for an agent with shell access
Connect Wyatt as a remote MCP server named "wyatt" using https://www.wyattworks.ai/api/mcp-server. Use Streamable HTTP and OAuth. Open the authorization flow for me, then verify the connection by listing the available Wyatt tools. Do not ask me to create or paste an API key.
Endpoint: https://www.wyattworks.ai/api/mcp-server — Streamable HTTP with browser OAuth and dynamic client registration.
What a connected agent can do
Search and read
Federated search across the workspace, plus list_documents and get_document with clean pagination, so an agent can pull the exact context it needs.
Write documents
create_document, update_document, and delete_document turn agent output into real workspace documents instead of chat scrollback.
Track tasks
list_tasks and get_task give the agent the current state of work, and get_workspace_context_next surfaces what to pick up.
Run background agents
kickoff_background_agent hands longer jobs to Wyatt itself, so your local agent can delegate and move on.
The connector path uses browser-based OAuth. Nothing on this page asks you to handle a secret.
You grant access in the browser through the Wyatt OAuth flow. Pasted API keys are rejected on purpose, so a leaked snippet never becomes a leaked credential.
On the consent screen you pick read-only or full access. Wyatt rejects any tool call that the granted scope does not allow.
Tokens are bound to the workspaces you approve. A connector for one workspace cannot read another.
An inactive connection expires on its own after 120 days.
No. Wyatt uses browser-based OAuth with dynamic client registration: your client opens a browser window, you sign in and approve access, and the client receives its own scoped token. Wyatt rejects pasted API keys on the MCP endpoint by design.
Anything that speaks MCP over Streamable HTTP: Claude Code, Codex, Cursor, ChatGPT (developer mode), OpenClaw, and other MCP-compatible clients. Claude Code additionally gets the Wyatt plugin with a /wyatt command and skills.
Thirteen tools: workspace overview and connection info, federated search, document list/read/create/update/delete, task listing and detail, next-action context, and kicking off Wyatt background agents. What the agent can use depends on the access level you grant.
The consent screen offers read-only or full access, bound to the workspaces you choose. Inactive connections expire after 120 days on their own.
They connect directly to Wyatt’s remote MCP server using the setup on this page. The plugin, /wyatt command, and bundled skills are additional Claude Code features.
The CLI is a private developer preview without a public installer. Use the MCP setup instead, and do not install third-party packages claiming to be the Wyatt CLI.