- Erste Schritte
- Best Practices
- Organisationsmodellierung im Orchestrator
- Beste Praktiken für die Automatisierung (Automation Best Practices)
- Optimieren von Unattended-Infrastruktur mithilfe von Maschinenvorlagen
- Organisieren von Ressourcen mit Tags
- Exportieren von Rastern im Hintergrund
- Durchsetzung der Governance der Integration Service-Verbindung auf Benutzerebene
- Mandant
- Über den Kontext „Mandant“
- Suche nach Ressourcen in einem Mandanten
- Verwaltung von Robotern
- Verbindung von Robotern mit Orchestrator
- Speicherung von Roboterzugangsdaten in CyberArk
- Speichern der Kennwörter von Unattended-Robotern im Azure Key Vault (schreibgeschützt)
- Speichern der Anmeldeinformationen von Unattended-Robotern im HashiCorp Vault (schreibgeschützt)
- Speichern der Anmeldeinformationen von Unattended-Robotern im AWS Secrets Manager (schreibgeschützt)
- Löschen von getrennten und nicht reagierenden Unattended-Sitzungen
- Roboter-Authentifizierung
- Roboter-Authentifizierung mit Client-Anmeldeinformationen
- Konfigurieren von Automatisierungsfunktionen
- Solutions (Lösungen)
- Audit
- Einstellungen
- Registrierung
- Cloud Robots
- Übersicht über Cloud Robots
- Ausführen von Unattended-Automatisierungen mit Cloud Robot – VM
- Hochladen Ihres eigenen Image
- Wiederverwenden von benutzerdefinierten Maschinen-Images (für manuelle Pools)
- Zurücksetzen der Anmeldeinformationen für eine Maschine (für manuelle Pools)
- Überwachung
- Sicherheitsupdates
- Testversion anfordern
- Häufig gestellte Fragen
- Konfigurieren einer VPN für Cloud-Roboter
- Konfigurieren einer ExpressRoute-Verbindung
- Live-Streaming und Remotesteuerung
- Events
- Anzeigen und Zugreifen auf Benachrichtigungen
- Anzeigen und Zugreifen auf E-Mail-Benachrichtigungen
- Es werden nur ungelesene Benachrichtigungen angezeigt
- Alle Benachrichtigungen als gelesen markieren
- Alle Benachrichtigungen löschen
- Löschen von Benachrichtigungen
- Abonnieren von Ereignissen
- Abbestellen von Ereignissen
- Automation Suite-Roboter
- Ordnerkontext
- Prozesse
- Jobs
- Apps
- Auslöser
- Protokolle
- Überwachung
- Indizes
- Warteschlangen
- Assets
- Über Assets
- Verwalten von Assets in Orchestrator
- Verwalten von Assets in Studio
- Speichern von Assets im Azure Key Vault (schreibgeschützt)
- Speichern von Assets im HashiCorp Vault (schreibgeschützt)
- Speichern von Assets im AWS Secrets Manager (schreibgeschützt)
- Speichern von Assets in Google Secret Manager (schreibgeschützt)
- Verbindungen
- Geschäftsregeln
- Speicher-Buckets
- MCP-Server
- Über MCP-Server
- MCP use cases and flows
- Testing MCP Servers
- Troubleshooting MCP Servers
- MCP-Compliance-Richtlinien
- Testverfahren in Orchestrator
- Ressourcenkatalogdienst
- Integrationen
- Fehlersuche und ‑behebung
Setup steps for connecting a public or private external tool to UiPath as a Remote MCP Server, and the authentication required for each.
Many of the tools your agents need already exist somewhere else, maybe in a software running quietly inside your own infrastructure. A Remote MCP Server lets you bring any of these into UiPath as an MCP tool, so your agents can discover and call them the same way they use tools built natively on the platform and inheriting the UiPath platform governance features.
How you connect depends on where the tool lives:
- Directly, if it's already public on the internet.
- Through UiPath Relay, if it lives inside your own network.
This page walks through both, with the setup steps for each. For the other MCP Server types, UiPath, Coded, Command, and Self-Hosted, see MCP Server types.
Anwendungsfälle
- Bring a partner's API into your automations: your team already has API access to a shipping carrier or a payment provider. As a Remote MCP Server, it's available to your agents directly, with no custom integration to build or maintain.
- Connect a SaaS tool your agents already need: many popular platforms publish their own public MCP Server. Once it's added in Orchestrator, it's available to every agent and automation with access to that folder.
- Reach an internal system without exposing it to the internet: your ticketing system or a legacy service runs inside your own data center and can't be published publicly. Through Relay, your agents can call its tools without a single inbound port opened.
Auf einen Blick
| Use Case 1: Relay | Use Case 2: Direct | |
|---|---|---|
| Am besten geeignet für | Tools inside your own network or data center | Tools already reachable on the internet |
| Extra setup | Relay Client installed and registered in your network | None beyond adding the MCP Server |
| Inbound firewall changes | None, Relay keeps an outbound-only tunnel open | Keine |
Use Case 1: External MCP via Relay + OAuth
Request flow
The scenario
A tool your agents need, an ERP, a ticketing system, or a legacy service, runs inside your own network, behind a firewall, and isn't reachable from the public internet. You want your UiPath agents to call its tools the same way they call any other MCP Server in your catalog, without exposing it to the internet.
Reachability and identity need to be addressed:
- Reaching a host with no public route - solved via UiPath Relay
- Proving the caller's UiPath identity - solved via OAuth dynamic authentication
The tool's own credential stays separate from both.
If the tool is already reachable from the public internet, use Use Case 2 instead as you don't need Relay.
Setting up the connection
Voraussetzungen:
- Relay provisioned for your tenant, with the Relay Client installed and registered in your network. See the Relay admin guide for enablement.
- Your identity has the
MCPServers.Viewpermission in the folder that will hold the MCP Server. The Automation User and Automation Developer roles include it. - The credential the tool itself requires, an API key or an Integration Service connection, is ready in that folder.
Anweisungen
- On the MCP Servers page, select Add MCP Server.
- Select the Remote type.
- Enter a Name for the MCP Server.
- Add a Description.
- Set the connection type to Private (Relay).
- Configure Authentication:
- Connection: select or add an Integration Service connection to be used for getting the authentication token.
- Authentication Token: in the Header section, add a static authentication token. We recommend referencing an asset instead of hardcoding a secret, for example
Authorization: %ASSETS/RemoteBearerToken%.
- In Remote URL, enter the tool's address as seen inside your network, the same address the Relay Client already reaches.
- Wählen Sie Hinzufügen aus.
Authenticating the caller
Callers authenticate the same way as with any other MCP Server: through the MCP OAuth flow for interactive clients like an IDE, or a personal access token, external application, or interactive login for automated callers. See MCP Server authentication for the full method matrix.
POST https://cloud.uipath.com/{org}/{tenant}/agenthub_/mcp/{folderKey}/{slug}
Authorization: Bearer <token>
Content-Type: application/json
{ "jsonrpc": "2.0", "method": "tools/list", "id": 1 }
POST https://cloud.uipath.com/{org}/{tenant}/agenthub_/mcp/{folderKey}/{slug}
Authorization: Bearer <token>
Content-Type: application/json
{ "jsonrpc": "2.0", "method": "tools/list", "id": 1 }
Your sign-in credentials are never shared with the tool. Its own credential, from the Headers or Connection field above, is applied separately on every call.
Verifying the connection
A tools/list call that returns the tool's tools confirms both the caller's identity and the tool's own credential passed. If the call fails:
- 401 usually means the caller's token.
- 403 usually means the folder is missing
MCPServers.View. - 502 or 504 usually means the Relay Client is offline, or the tool rejected its own credential.
See Troubleshooting MCP Server authentication and Troubleshooting MCP Servers for more detail.
Use Case 2: External MCP, direct + OAuth
The scenario
A tool your agents need is already published on the public internet, a SaaS or partner MCP Server, or one you host and expose yourself. You want your agents to call it the same way they call any other MCP Server, with the same governance and audit trail, without needing Relay.
If the tool is private or on-premises, use Use Case 1 instead.
Setting up the connection
Voraussetzungen:
- Your identity has the
MCPServers.Viewpermission in the folder that will hold the MCP Server. The Automation User and Automation Developer roles include it. - The credential the tool itself requires, an API key or an Integration Service connection, is ready in that folder.
Anweisungen
- On the MCP Servers page, select Add MCP Server.
- Select the Remote type.
- Enter a Name for the MCP Server.
- Add a Description.
- Set the connection type to Standard.
- Configure Authentication:
- Connection: select or add an Integration Service connection to be used for getting the authentication token.
- Authentication Token: in the Header section, add a static authentication token. We recommend referencing an asset instead of hardcoding a secret, for example
Authorization: %ASSETS/RemoteBearerToken%.
- In Remote URL, enter the tool's address as seen inside your network.
- Wählen Sie Hinzufügen aus.
Authenticating the caller
Callers authenticate the same way as with any other MCP Server: through the MCP OAuth flow for interactive clients like an IDE, or a personal access token, external application, or interactive login for automated callers. See MCP Server authentication for the full method matrix.
POST https://cloud.uipath.com/{org}/{tenant}/agenthub_/mcp/{folderKey}/{slug}
Authorization: Bearer <token>
Content-Type: application/json
{ "jsonrpc": "2.0", "method": "tools/list", "id": 1 }
POST https://cloud.uipath.com/{org}/{tenant}/agenthub_/mcp/{folderKey}/{slug}
Authorization: Bearer <token>
Content-Type: application/json
{ "jsonrpc": "2.0", "method": "tools/list", "id": 1 }
Verifying the connection
A tools/list call that returns the tool's tools confirms the connection works. If the call fails:
- 401 usually means the caller's token.
- 403 usually means the folder is missing
MCPServers.View. - 502 usually means the tool is unreachable, or it rejected its own credential.
- A connection error where none is expected usually means the address resolves to a private or internal host; switch to Use Case 1 instead.
See Troubleshooting MCP Server authentication and Troubleshooting MCP Servers for more detail.
- Anwendungsfälle
- Auf einen Blick
- Use Case 1: External MCP via Relay + OAuth
- Request flow
- The scenario
- Setting up the connection
- Authenticating the caller
- Verifying the connection
- Use Case 2: External MCP, direct + OAuth
- The scenario
- Setting up the connection
- Authenticating the caller
- Verifying the connection