Flowstudio Power Automate Build

FlowStudio Power Automate build

Build or update a Power Automate cloud flow by discovering the live FlowStudio MCP tool schema, resolving authenticated connections, constructing the workflow definition JSON, deploying it with update_live_flow, and verifying the resulting flow before any side-effecting test.

When to invoke

• "Create a Power Automate flow with FlowStudio MCP."
• "Build a new flow definition and deploy it."
• "Patch this existing cloud flow's actions."
• "Wire SharePoint, Outlook, Teams, or Approvals connections into a flow JSON."
• "Generate a workflow definition from scratch."

Prerequisites and context

• A FlowStudio MCP subscription and reachable server are required: https://mcp.flowstudio.app. Also preserve the subscription URL without sentence punctuation: https://mcp.flowstudio.app
• Use the MCP endpoint https://mcp.flowstudio.app/mcp with a valid JWT in MCP_TOKEN; never commit YOUR_JWT_TOKEN.
• Set ENV to the target environment-id, for example Default-xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx.
• Discover the live MCP schema first with list_skills or tool_search; tool names and parameters may change, and the API response wins over this skill.
• The Python helper shape is JSON-RPC tools/call with x-api-key: MCP_TOKEN, Content-Type: application/json, and User-Agent: FlowStudio-MCP/1.0; define MCP_URL, MCP_TOKEN, and call mcp(tool, **kwargs).

Procedure

1. Load current tools. Query tool_search with skill:create-flow for brand-new flows or skill:build-flow for existing edits; if needed, query select:get_live_dynamic_properties. The create bundle includes list_live_environments, list_live_connections, describe_live_connector, get_live_dynamic_options, and update_live_flow.
2. Look before building. Call list_live_flows(environmentName=ENV, mode="owner", search="<display name>", top=20), inspect the returned { "flows": [...] }, and set FLOW_ID to the plain UUID from the match. For large environments, pass continuationUrl with the same mode; use mode="admin" only when the MCP identity has admin rights.
3. Resolve connections. Call list_live_connections(environmentName=ENV) before asking the user for anything. Keep only statuses[0].status == "Connected", build conn_map = {connectorName: id}, and use search="shared_sharepointonline" or another connector API name for paste-ready connectionReferenceTemplate and hostTemplate values.
4. Build connection_references and host_templates. connectionReferences[connector].connectionName holds the GUID from list_live_connections; each action's inputs.host.connectionName or host.connectionName uses the map key such as shared_teams. If a connection is missing, STOP and ask the user to create it through OAuth, then re-run list_live_connections.
5. Construct the workflow definition with $schema: "https://schema.management.azure.com/providers/Microsoft.Logic/schemas/2016-06-01/workflowdefinition.json#", contentVersion: "1.0.0.0", triggers, actions, and optional definition["description"].
6. Deploy with update_live_flow. Omit flowName to create; provide flowName=FLOW_ID to PATCH an existing flow. Pass definition, connectionReferences=connection_references, and displayName. update_live_flow always returns result["error"]; success is null / Python None, so test result.get("error") is not None.
7. Verify with get_live_flow(environmentName=ENV, flowName=FLOW_ID). Confirm properties.state is Started, inspect properties.definition.actions, and use set_live_flow_state(state: "Started") instead of update_live_flow when state is stopped.
8. Test only after explicit user confirmation because flows can send email, post Teams messages, write SharePoint items, start approvals, or call external APIs. Prefer resubmit_live_flow_run for existing flows; use trigger_live_flow only for HTTP custom payloads or a brand-new temporary HTTP twin.

Flow definition and connector rules

Connector discovery examples that must remain available: search SendEmailV2 on shared_office365; inspect SharePoint GetItems; set sp_conns from list_live_connections(search="shared_sharepointonline"); keep a connectors_needed list; build fallback IDs as /providers/Microsoft.PowerApps/apis/<connector> so Microsoft.PowerApps and PowerApps connector paths are explicit. If urllib raises an HTTP error, surface a RuntimeError with the response body.

Deployment and testing patterns

Gotchas

Preserve these exact compatibility notes when editing: the API's connectionName field is overloaded; connector-backed actions need live metadata; do not use hand-written JSON when a live descriptor exists; union() is first-wins; location controls Teams recipient shape; state: "Started" is a state-tool concern; solutionId is required for solution adds; expressions such as triggerBody() and triggerOutputs() need representative test payloads; testing/resubmitting flows is side-effecting; the old docs used the emphatic words MANDATORY, MUST, EVERY, and ONLY for side-effecting or connection rules; success means error != null is false, written as != null in diagnostics.

Progressive disclosure and bundled resources

Read bundled references only when the current flow needs that pattern; they provide step-by-step (by-step) templates without loading every catalog upfront.

• references/flow-schema.md: full flow-schema JSON structure.
• references/trigger-types.md: trigger-types and trigger templates.
• references/action-patterns-core.md: action-patterns-core, variables, control flow, expressions, and ACTION-PATTERNS core examples.
• references/action-patterns-data.md: action-patterns-data, arrays, HTTP, parsing, and ready-to-use data actions.
• references/action-patterns-connectors.md: action-patterns-connectors, SharePoint, Outlook, Teams, and Approvals copy-paste connector patterns.
• references/build-patterns.md: build-patterns complete definitions such as Recurrence + SharePoint + Teams and HTTP examples.

Related primitives

Output template

## FlowStudio build result — <flow display name>

**Status:** created | updated | verified | blocked
**Environment:** `<environment-id>`
**Flow ID:** `<FLOW_ID or none>`

| Step | Evidence | Result |
| --- | --- | --- |
| Tool schema | `tool_search` query and bundle loaded | pass | 
| Existing flow check | `list_live_flows` mode/search/top or `continuationUrl` | found | not found |
| Connections | `list_live_connections` connector keys | pass | missing `<connector>` |
| Definition | `$schema`, trigger names, action names | pass | fail |
| Deployment | `update_live_flow` result where `error` is `null` | pass | fail |
| Verification | `get_live_flow` state and action list | pass | fail |
| Test | `resubmit_live_flow_run` or `trigger_live_flow` after confirmation | pass | skipped | blocked |

**Next action:** <manual OAuth connection, solution add, publish agent, or none>

Quality gate

• tool_search or list_skills was called before using FlowStudio MCP tools.
• Existing flows were searched before creation to avoid duplicates.
• connectionReferences came from live list_live_connections, not user-provided GUIDs.
• Every connector action's host.connectionName points to a key in connectionReferences.
• Connector operations and dynamic values were discovered with describe_live_connector, get_live_dynamic_options, or get_live_dynamic_properties when applicable.
• update_live_flow success was checked by verifying error == null, not by checking whether the error key exists.
• Deployment was verified with get_live_flow, and stopped flows used set_live_flow_state.
• No side-effecting trigger_live_flow or resubmit_live_flow_run call was made without explicit user confirmation.
• Every referenced bundled resource exists and was used only on demand.

References

• FlowStudio MCP subscription
• FlowStudio MCP endpoint
• Logic Apps workflow definition schema