Overview
Your Taskade AI agents can reach out to external tools by connecting to other companies' MCP servers. Notion, Linear, and any service that exposes a Model Context Protocol endpoint becomes a tool your agent can call. The connection is one-time, the agent uses it forever.
TL;DR: Taskade plays both sides of the Model Context Protocol. Other AI clients can connect to Taskade as a server (covered in MCP Server). And your Taskade agents can connect out to external MCP servers like Notion or Linear. This article covers the outbound direction. One connector, every agent in the workspace can use it. Build a custom agent →.
How it works
You add an MCP server to your workspace once. After that, any custom agent you build can be allowed to use it. The agent sees the server's tools the same way it sees Taskade's built-in tools.
A single connector handles auth and tool listing for every agent that needs it. No per-agent setup once the workspace has the server.
Add an external MCP server
You add the server in your workspace settings, then pick which agents can use it.
- Open Settings in your workspace.
- Click Connectors.
- Click Add connector and pick MCP Server.
- Paste the server's URL and sign in if the server requires it.
- Save. The connector is now available across the workspace.
Settings › Connectors — each MCP server shows ● Active or ○ Needs Auth, and one connector serves every agent in the workspace:
┌────────────────────────────────────────────────────────────┐
│ Settings › Connectors [ + Add Connector ]│
├────────────────────────────────────────────────────────────┤
│ ● Notion v1.2 [Active] │
│ Read and search your team wiki │
│ Updated Jun 1 · key: notion-workspace │
│ │
│ ○ Linear [Needs Auth] [ Connect ] │
│ Open issues, set priority, assign owners │
│ │
│ ● Sentry v0.9 [Active] │
│ Pull error details into triage chat │
│ │
│ One connector — every agent in the workspace can use it. │
└────────────────────────────────────────────────────────────┘
Let an agent use the server
Once the connector is in place, allow a specific agent to call it.
- Open the agent's settings.
- Click Tools.
- Enable the MCP server in the tool list.
- Save.
The agent now sees every tool the server exposes. Ask the agent a question that calls for one of those tools, and it will pick up the right one automatically.
What this unlocks
- A Notion-aware research agent that can read your team's wiki without you exporting anything.
- A Linear-aware triage agent that can open issues, set priority, and assign owners straight from chat.
- A multi-server agent that can pull a fact from one server, reason on it, then write a result to another.
Plan availability
MCP server access (both inbound to Taskade and outbound from your agents) is part of Business, Max, and Enterprise plans. The full plan matrix lives at Pricing.
Tips
- One server, many agents. Add the connector once and reuse it. Avoid wiring the same server into every agent's prompt by hand.
- Tool descriptions matter. External MCP servers ship with their own tool docs. The clearer those docs, the better your agent picks the right tool. If a server's docs are vague, write a one-line description on the connector that explains when to reach for it.
- Watch the credit cost. Calling an external server uses a tool-call credit like any other tool. The model picker shows the live cost. See AI Models & Credit Costs.
Frequently Asked Questions
Which clients and servers can my agents connect to?
Your Taskade AI agents can connect to any service that exposes a Model Context Protocol endpoint. Notion and Linear are common picks, and any MCP-compatible server works the same way. You add the server's URL once in your workspace settings, and every agent you allow can use it.
How do I connect an external MCP server?
Open Settings in your workspace, click Connectors, then Add connector and pick MCP Server. Paste the server's URL and sign in if the server asks you to. Save, and the connector is ready for any agent in the workspace to use.
Which plan do I need to use MCP servers?
MCP server access is part of the Business, Max, and Enterprise plans. This covers both directions: external servers your agents reach out to, and other AI clients that connect into Taskade. See Pricing for the full plan matrix.
Do I need an API key to connect a server?
It depends on the server. Many external MCP servers use a point-and-click sign-in, so you just sign in when you add the connector and no key is required. If a server asks for credentials, the Add connector flow walks you through it.
Do I have to set up the server for every agent?
No. You add the MCP server once at the workspace level, then allow specific agents to use it from each agent's Tools settings. One connector serves every agent you choose, so there is no per-agent wiring beyond flipping the tool on.
How can other AI tools connect into Taskade instead?
That is the inbound direction, where clients like Claude Desktop, Cursor, Windsurf, and Zed connect to Taskade through the hosted endpoint with a point-and-click sign-in and no API key. See Taskade as an MCP Server for the setup steps and plan details.
Does calling an external server cost credits?
Yes. When an agent calls a tool on an external MCP server, it uses a tool-call credit like any other tool. The model picker shows the live cost before you run it. See AI Models & Credit Costs for the details.
Related guides
- Taskade as an MCP Server. The other direction. Let Claude Desktop or Cursor query your workspace.
- Custom AI Agents. Where you allow which agents to use which servers.
- Agent Tools. All the tool types an agent can call, MCP included.
- Agent Knowledge & Memory. Knowledge files complement MCP tools for full-context agents.
- AI Models & Credit Costs. Live cost on every tool call.