Skip to main content

​
Bring the MCP ecosystem into MagOneAI

MagOneAI ships with first-party integrations (Google, Microsoft, databases, web search, and more — see the tools overview). But MCP is an open protocol, and a growing ecosystem of third-party MCP servers already exposes tools for CRMs, ticketing systems, analytics platforms, internal APIs, and more. External MCP servers let you connect any of those to your organization by pasting a URL. The tools the server exposes become available to agents in your projects immediately — no image to build, no integration to write.
If you’re new to MCP, start with the MCP protocol overview — it explains why MagOneAI uses MCP and how first-party integrations work.

​
When to use external MCP

Third-party MCP server

A vendor or open-source project already provides an MCP server for the service you want to integrate. Connect by URL.

Your own hosted MCP server

You’ve built a custom MCP server internally and deployed it. Point MagOneAI at the URL and your agents can use it.

Prototyping and evaluation

Trying out a new MCP-compatible tool before committing to a deeper integration.

Org-specific integrations

You need a tool for one organization but not platform-wide. External MCP keeps it scoped to that org.
For tools you need to customize deeply or ship alongside the platform, consider building a custom MCP server instead.

​
Connecting an external server

Only organization owners can add external MCP servers, because tools connected this way are shared across every project in the org.
1

Open Tools → Add external MCP

From the organization’s tool settings, choose to add an external server.
2

Provide the server URL

Paste the HTTPS URL of the MCP server. The URL must be publicly reachable — private IPs and internal-only hostnames are rejected for security (see below).
3

Configure authentication headers

Most MCP servers require authentication. Add the headers the server expects — for example, an Authorization header with a bearer token, or a vendor-specific API key header. Credential values are stored securely and never appear in workflow logs.
4

Discovery

MagOneAI calls the server to fetch its tool catalog. Each tool’s name, description, and parameter schema is imported automatically. If the server is unreachable or the credentials are wrong, you’ll see a clear error at this step rather than silent failure later.
5

Use in workflows

Once discovery succeeds, the new tools show up in the tool picker on the canvas. Agents in any project in this organization can be given access to them.

​
Security considerations

External MCP URLs must use https:// and must resolve to a public address. Private IPs, localhost, and internal-only hostnames are blocked to prevent the platform from being used to reach networks it shouldn’t. If you need to connect an internal-only server, host it behind a public HTTPS endpoint (with its own auth) or package it as a custom MCP server deployed alongside MagOneAI.
The credential values you enter are sent directly to the platform’s secure vault. Workflows reference them indirectly, and they’re only injected into outbound requests to the MCP server at runtime — never logged, never exposed in exports.
Some headers are reserved by the transport (Host, Content-Type, Content-Length, and similar) and cannot be used for authentication. MagOneAI rejects these at configuration time so you notice during setup, not during a live run.
External MCP servers are scoped to the organization that added them. Other organizations on the same MagOneAI deployment can’t see or use them.

​
Ongoing management

Once a server is connected, you can manage it from the same tool settings page:
  • Re-discover tools — if the upstream server adds or removes tools, trigger a refresh to pull the latest catalog.
  • Rotate credentials — update the stored credential values in place without reconnecting the whole server.
  • Rotate the organization’s service token — MagOneAI authenticates to the upstream connector on your behalf using a per-org service token; rotate it if you suspect it’s been compromised.
  • Disconnect — remove the server when it’s no longer needed. Workflows that reference its tools will show a clear “tool unavailable” error rather than silently skipping.

​
What external MCP doesn’t do

  • It doesn’t host the server for you. The MCP server itself runs somewhere else — wherever the vendor or your own infrastructure hosts it. MagOneAI is the client.
  • It doesn’t modify the tool contract. The tools’ names, descriptions, and parameter schemas come from the upstream server. If you need to rename or reshape a tool, you’d do that on the server side.
  • It doesn’t replace first-party integrations. For heavily used services (Google, Microsoft, databases), the built-in integrations offer deeper platform features — OAuth flows, per-user scoping, richer credential management. Use external MCP when those integrations don’t already cover what you need.

​
Next steps

MCP protocol

Understand how MCP works and why MagOneAI uses it for tool integration

Custom tools

Build your own MCP server when external connection isn’t enough

Tools overview

See the full catalog of built-in integrations

Secrets management

How MagOneAI stores and injects credentials securely