Skip to main content
An MCP instance is your project’s deployment of an MCP server from the platform catalog (or a custom server you register). Instances make tools available to workflow mcp_call steps. Manage them from MCP (/mcp) in the Console. The page title is MCP connectors.

​
Catalog vs instance

TermMeaning
MCP server (catalog)Platform-wide definition — adapter slug, tool list, config schema
MCP instanceYour project’s installed copy with config profiles, env, and connection wiring
The catalog lists 40+ Go adapters (Gmail, Postgres, Shopify, GitHub, …). See the connector catalog.

​
Installing an instance

1

Open MCP connectors

Go to MCP in the sidebar (/mcp). Browse Platform catalog servers or your installed instances.
2

Add an instance

On a catalog server row, click Add instance (or open the platform marketplace modal). First-party platform servers create a workspace instance you own.
3

Open instance config

Open Instance config (/mcp/instances/{instance_id}). Description: “Pick a profile, connect credentials, then save. Workflows use the profile you mark active.”
4

Wire a connection

On the active config profile, select a connection in the credential / source_spec wiring. For Google and WhatsApp, use the embedded connect flows when offered.
5

Set active profile

Mark the profile active so runtime resolution uses it for this instance.
6

Validate

Optionally run MCP validation at /mcp/{server_id}/validate to check server-level health, or validate a connection binding via API.

​
Using instances in workflows

In Workflow Studio (/workflow/{id}), add an mcp_call step and select:
  • The instance (saved as instance_id + server_url on the step)
  • The tool name from the instance’s tool catalog
  • tool_args — static values or template variables ({{input.*}}, {{steps.step-id.result.*}})
At runtime, the workflow engine calls the MCP tool with credentials from the instance’s active config profile.

​
Importing shared graphs

When you Import a workflow JSON from another workspace or a template:
  1. Studio previews the graph (fit-to-screen).
  2. MCP steps grouped by server_url show connection status for your workspace.
  3. Pick an existing instance, Create instance, or connect credentials for servers you have not installed yet.
  4. Unknown servers must be added from MCP before import can complete.
Shared graphs should include server_url (portable). instance_id values from the source workspace are remapped on import.

​
Connection overrides (API)

Workflow-level overrides (PUT /v1/workflows/{id}/connection-overrides) let one graph target different credentials per environment. Configure via API today; see API reference.

​
Custom MCP servers

Register tenant-owned MCP servers you operate (MCP → Add connectors → Add connector) and deploy instances the same way as catalog adapters. See the Custom MCP servers guide for the full wizard, API flow, and troubleshooting.

​
Platform MCP (IDE integration)

Separate from project instances, Platform MCP at mcp.agentruntime.io exposes Console APIs as tools for IDEs and assistants. Authenticate with a PAT. Use it to list workflows, start runs, and manage resources from Cursor or other MCP clients.