MCP servers
Connect an MCP server. Then just say what you want done.
The Model Context Protocol lets an AI agent use real tools instead of describing them. Ottermind hosts the connection, so there is no local process, no JSON config to hand-edit, and no restart when something changes. Pick a server below, or paste any MCP URL.
Servers you can connect today
Each one is a working connection in Ottermind, not a directory listing.
- Playwright MCP
Playwright MCP drives a real browser from an AI agent using accessibility snapshots instead of pixel guessing. Ottermind ships the same model as built-in tools, so a browser task works in a fresh chat with nothing to install.
BrowserBuilt in - GitHub MCP server
A GitHub MCP server lets an agent read repositories, issues and pull requests instead of pasting code into chat. Ottermind connects GitHub over OAuth, so the agent works against your real repos with the scopes you approve.
DeveloperOAuth connector - Notion MCP
Notion MCP gives an agent search and write access to your workspace. Ottermind connects Notion over OAuth and keeps the pages it reads as task context, so follow-up questions do not need the same links again.
Docs & knowledgeOAuth connector - Linear MCP
Linear MCP lets an agent read and update issues, projects and cycles. Ottermind connects Linear over OAuth with read access by default — write scopes stay off until you grant them.
DeveloperOAuth connector - Supabase MCP
Supabase MCP exposes your project's schema and data to an agent so it can answer questions in SQL rather than guesswork. Ottermind treats it as a managed connector — no local process, no config file to maintain.
DatabaseManaged MCP - Slack MCP server
A Slack MCP server lets an agent search conversations and post updates. Ottermind connects Slack over OAuth with search and channel reads on by default, and keeps posting behind an explicit write scope.
Team chatOAuth connector - Sentry MCP
Sentry MCP puts your error stream in reach of an agent, so a stack trace can be read, grouped and explained without leaving the task. Ottermind connects it as a managed connector.
MonitoringManaged MCP - Stripe MCP
Stripe MCP lets an agent read customers, subscriptions, invoices and payment events. Ottermind connects it as a managed connector, and any write operation asks for confirmation before it runs.
PaymentsManaged MCP - Gmail MCP
Gmail MCP gives an agent search and drafting access to your inbox. Ottermind connects Gmail over OAuth with read-only scope by default; sending and drafting are separate grants you turn on deliberately.
EmailOAuth connector - Vercel MCP
Vercel MCP lets an agent read projects, deployments and build logs. Ottermind connects it as a managed connector, so a failed deploy can be diagnosed in the same task where you fix it.
DeveloperManaged MCP
Bring your own MCP server
Paste any MCP endpoint into Ottermind and it is checked, listed and ready in the same conversation. Servers you add stay private to your workspace.
Add a custom MCP serverMCP questions people actually ask
What is an MCP server?
An MCP server is a small program that exposes one system's abilities — a database, an inbox, a browser — through the Model Context Protocol, a shared format AI agents understand. Instead of writing a custom integration per model, you run one server and any MCP-capable agent can use it.
What is MCP in AI?
MCP stands for Model Context Protocol. It standardizes how an AI model discovers and calls external tools, and how those tools return results. It is the reason an agent can search your Notion, read a Sentry trace and open a browser tab without three bespoke integrations.
MCP vs API — what's the difference?
An API is built for a programmer who reads documentation first. MCP is built for a model: the server describes its own tools, arguments and results at runtime, so the agent can discover what is available and choose. MCP servers usually sit on top of an existing API rather than replacing it.
Which MCP servers does Ottermind support?
The list on this page reflects what is connectable today, including Playwright-style browser control, GitHub, Notion, Linear, Supabase, Slack, Sentry, Stripe, Gmail and Vercel. You can also add any other MCP server by URL.
Do I need to run an MCP server locally?
No. Ottermind hosts the connection for the servers listed here, so there is no local process to keep alive and no config file to maintain. If you do want a local server, the desktop app can reach one running on your own machine.
Can I add my own MCP server?
Yes. Paste the server URL and Ottermind checks it, lists its tools and makes it available in the conversation. Servers you add are private to your workspace, and any tool that writes asks for confirmation before it runs.
What can the agent actually access?
Only what you grant. OAuth connections start read-only and keep write scopes separate, managed MCP connections are scoped per workspace, and write operations require an explicit confirmation. You can review or revoke any connection at any time.
Does this work with Claude?
Yes. Ottermind runs on Claude models and speaks MCP, so a connected server is available to the agent in every task. You do not need to edit a desktop config file to make a server visible.
Stop configuring. Start with the task.
Connect what you use, describe the outcome, and get a result you can edit — not a transcript of tool calls.
Try Ottermind free