Install Flow Studio MCP for Power Automate

What is Flow Studio MCP?

Flow Studio MCP is an MCP server for Power Automate, a cloud service that lets AI agents build, debug, deploy, and monitor your flows without you opening the Power Automate portal.

You talk to your AI agent in natural language. The agent talks to Flow Studio MCP. Flow Studio MCP talks to the Power Platform APIs on your behalf.

What you can do

Once connected, you can ask your agent to do things like:

For the full list of 30+ tools available, see MCP Tools Reference.

What you need

• A Microsoft account with access to Power Automate environments
• An MCP-compatible AI agent, including but not limited to Copilot Studio Agent, GitHub Copilot, Claude and Codex.
• A Flow Studio MCP account (free evaluation, 100 calls over 21 days, no credit card)

Step 1, Subscribe and get your API key

1a. Sign up and activate

Go to mcp.flowstudio.app and click Get Started. Sign in with your Microsoft account.

On the Dashboard, click Start Free, 100 calls to activate the free evaluation. No credit card required. You get 100 API calls over 21 days.

1b. Grant Power Platform permissions

Flow Studio needs access to your Power Automate and Power Apps services. This uses Microsoft’s standard OAuth consent, we never see your password.

1. Click Connect next to Power Automate, you will be redirected to Microsoft to authorize
2. After Power Automate is connected, click Continue setup next to Power Apps to complete the second authorization

Both services must show Connected before you can generate your API key.

1c. Copy your API key

Once both permissions are connected, your API key appears in the Your API Key section. Click the copy button to copy it. You will need this for Step 2.

The dashboard also shows your MCP endpoint (https://mcp.flowstudio.app/mcp) and a Quick start JSON config you can copy directly into your agent.

Step 2, Configure your agent

Your API key is a JWT token. The setup is different for each agent. The fastest options are listed first:

For manual configuration or other agents, follow the guide for the agent you use:

Microsoft Copilot Studio for Power Automate

Wire Flow Studio MCP into a Copilot Studio agent and any M365 Copilot user in your tenant can use it from Teams chat or m365.cloud.microsoft, with no install on their side.

1. Open copilotstudio.microsoft.com, create a new agent (or open an existing one), and go to the Tools tab.
2. Click + Add a tool → in the Create new row, click the Model Context Protocol card.
3. Fill the form: Server URL https://mcp.flowstudio.app/mcp, Authentication = API key, Type = Header, Header name = x-api-key (lowercase, with the hyphen, this is the most common setup mistake). Click Create.

Set up GitHub Copilot in VS Code

Full guide: GitHub Copilot for Power Automate with Flow Studio MCP

Recommended: Install the Power Automate MCP Server extension from the VS Code Marketplace. Open VS Code, press Ctrl+Shift+X, search for "Flow Studio MCP", and click Install. The extension walks you through adding your API key and stores it securely in your OS keychain (Windows Credential Manager / macOS Keychain). It supports multiple tenant connections and registers the MCP server with Copilot Chat automatically.

Alternative (manual JSON): Create or open .vscode/mcp.json in your project root and add:

{
  "servers": {
    "flowstudio": {
      "url": "https://mcp.flowstudio.app/mcp",
      "headers": {
        "x-api-key": "<your-api-key>"
      }
    }
  }
}

Replace <your-api-key> with the API key from your dashboard. VS Code will show Running with 30+ tools when the connection succeeds. To avoid committing your key to source control, add .vscode/mcp.json to your .gitignore, or use the global config via MCP: Open User Configuration command.

Optionally, install our pre-built Copilot agent skills for guided workflows. Copy the skill folders into your project’s .github/skills/ directory. See Copilot Agent Skills for details.

Set up Claude Code

Full guide: Claude for Power Automate with Flow Studio MCP

Claude Code runs in three places: inside an editor (VS Code, Cursor, VSCodium, Windsurf), in the terminal as the claude CLI, and as a Desktop app. All three read from the same user-scope config under ~/.claude/, so a single one-time setup configures the MCP connection everywhere. Three install paths, all first-class, pick whichever fits your workflow:

Option 1, VS Code extension (one-click, easiest if you already use VS Code / Cursor / VSCodium / Windsurf): Install the Power Automate MCP for Claude Code extension from the VS Code Marketplace. (Cursor, VSCodium, and Windsurf users: install from Open VSX.) The extension uses VS Code purely as the install interface, after setup, the connection works wherever Claude Code runs.

1. Install the extension from the VS Code Marketplace (or Open VSX mirror).
2. Get an API key at mcp.flowstudio.app, free to start, 100 calls included.
3. In VS Code, open the command palette and run Flow Studio (Claude): Add Tenant Connection. Enter a tenant name and paste your API key.
4. Reload the window. Claude Code in VS Code, in the terminal, and on Desktop all pick up the new MCP server and the 5 bundled skills (power-automate-mcp, power-automate-debug, power-automate-build, power-automate-monitoring, power-automate-governance).

The extension writes to ~/.claude/skills/ (skills) and ~/.claude.json (MCP server entry). It also adds Flow Studio (Claude): Remove and List Tenant Connections commands for managing multiple tenants.

Option 2, Plugin install via the Claude CLI (no editor required): Same 5 skills and same MCP connection as Option 1, installed from the terminal. Use this if you don't run a supported editor, or if you just prefer the CLI:

git clone https://github.com/ninihen1/power-automate-mcp-skills.git
claude --plugin-dir ./power-automate-mcp-skills

Then connect the MCP server:

claude mcp add --transport http flowstudio https://mcp.flowstudio.app/mcp \
  --header "x-api-key: <your-api-key>"

Option 3, Manual JSON config (MCP connection only, no skills): If you only want the MCP server connection (without the agent skills) or want full control over the config, edit Claude Code's config directly. Add the following to .mcp.json in your project root (or ~/.claude.json for user-level):

{
  "mcpServers": {
    "flowstudio": {
      "type": "http",
      "url": "https://mcp.flowstudio.app/mcp",
      "headers": {
        "x-api-key": "${FLOWSTUDIO_API_KEY}"
      }
    }
  }
}

Set the environment variable before running Claude Code:

export FLOWSTUDIO_API_KEY="<your-api-key>"

Claude Code supports ${VAR} syntax for environment variable expansion in headers.

Set up Claude Desktop

Full guide: Claude for Power Automate with Flow Studio MCP

Claude Desktop does not natively connect to remote HTTP MCP servers via config. You need the mcp-remote package as a bridge (requires Node.js).

Open your config file:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Add the following:

{
  "mcpServers": {
    "flowstudio": {
      "command": "npx",
      "args": [
        "mcp-remote@latest",
        "https://mcp.flowstudio.app/mcp",
        "--header",
        "x-api-key: YOUR_API_KEY",
        "--header",
        "User-Agent: FlowStudio-MCP/1.0"
      ]
    }
  }
}

Replace YOUR_API_KEY with the API key you copied from the dashboard. Restart Claude Desktop after saving the file. You can also add the server via Settings > Connectors > Add to avoid editing JSON.

OpenAI Codex for Power Automate

Codex runs as both a desktop app (UI configuration) and a CLI (config.toml). Both use the same MCP endpoint and require the x-api-key header.

Codex Desktop app (recommended, UI configuration): Open Settings > MCP servers > Add server and fill in the form:

• URL: https://mcp.flowstudio.app/mcp
• Bearer token env var: leave empty. Do not use this field, it sends an Authorization: Bearer header, which Flow Studio does not accept.
• Header, name x-api-key (lowercase, with the hyphen). Two ways to provide the value:
  • Direct: in the Headers section, add a row with Key x-api-key and paste your API key (the JWT) into the Value field.
  • Via env var (recommended): in the Headers from environment variables section, add a row with Key x-api-key and Value FLOWSTUDIO_API_KEY (or any env var name you choose). Then set that env var on your machine before launching Codex (export FLOWSTUDIO_API_KEY="<your-api-key>" on macOS/Linux, or the equivalent on Windows). This keeps the JWT out of the config UI and out of any synced settings.

Click Save. The Codex Desktop app will pick up the connection immediately, no app restart needed.

Codex CLI (config.toml): Add the following to your ~/.codex/config.toml (global) or .codex/config.toml (project):

[mcp_servers.flowstudio]
url = "https://mcp.flowstudio.app/mcp"
startup_timeout_sec = 10
tool_timeout_sec = 120

[mcp_servers.flowstudio.env_http_headers]
x-api-key = "FLOWSTUDIO_API_KEY"

Then set your API key as an environment variable:

export FLOWSTUDIO_API_KEY="<your-api-key>"

The env_http_headers section maps header names to environment variable names, Codex reads the value at runtime.

Important: Do not use bearer_token_env_var. That sends an Authorization: Bearer header, which Flow Studio does not accept. Use env_http_headers with x-api-key as shown above.

Codex network access (CLI): Codex CLI blocks network access by default. You need to enable internet access for your environment and add mcp.flowstudio.app to the domain allowlist. Without this, Codex CLI cannot reach the MCP server. (The Desktop app does not have this restriction by default.)

Other MCP-compatible agents for Power Automate

For agents that don’t have native MCP configuration (or for “simpler” agents), you can give them the connection details directly. Paste the following into your agent’s system prompt or first message:

Endpoint:  https://mcp.flowstudio.app/mcp
Protocol:  JSON-RPC 2.0 over HTTP POST
Transport: Streamable HTTP, single POST per request, no SSE, no WebSocket
Auth:      x-api-key header (NOT Authorization: Bearer)

Required headers:
  Content-Type: application/json
  x-api-key: <YOUR_API_KEY>
  User-Agent: FlowStudio-MCP/1.0

Step 1, Discover tools:
  POST {"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}
  → returns all tools with names, descriptions, and input schemas
  → free, not counted against plan limits

Step 2, Call a tool:
  POST {"jsonrpc":"2.0","id":1,"method":"tools/call",
        "params":{"name":"<tool_name>","arguments":{...}}}

Response shape:
  Success → {"result":{"content":[{"type":"text","text":"<JSON string>"}]}}
  Error   → {"result":{"content":[{"type":"text","text":"{\"error\":{...}}"}]}}

  Always parse result.content[0].text as JSON to get the actual data.

This approach works with any agent that can make HTTP requests, no MCP SDK required. It is the most reliable method for agents that do not natively support MCP configuration.

Important: The auth header is x-api-key, not Authorization: Bearer. This is the most common setup mistake.

Step 3, Try your first command

Once configured, start a conversation with your agent and try:

“List my Power Platform environments”

If the connection is working, you will see your environments with their names, regions, and SKUs. This confirms that your agent can reach Flow Studio MCP and your API key is valid.

Then try:

“List my flows in the default environment”

Power Platform organizes resources into environments. Most users have a default environment where their flows live.

“Show me the last 5 runs of [flow name]”

“What errors occurred in my most recent failed flow run?”

How to talk to your agent

You do not need to know tool names or API details. Just describe what you want in plain language. Here are some tips:

Be specific about which flow. Say the flow name: “Check the runs of my Invoice Processing flow” rather than “Check my flows.”

Ask follow-up questions. After seeing a failed run, ask “What went wrong?” or “Show me the error details.” Your agent remembers the context within a conversation.

Let the agent handle the technical parts. You do not need to provide environment IDs, flow GUIDs, or run IDs. The agent discovers these automatically. Just use display names.

Tell the agent what you want to achieve, not which tool to call. Say “Fix the expression error and redeploy” rather than “Call update_live_flow.”

Ask for explanations. After the agent finds an error, ask “What does this error mean?” or “How should I fix this?” The agent can explain Power Automate concepts and suggest fixes.

Troubleshooting

If your first call does not work, check these common issues:

Learn more: Flow Studio MCP on Microsoft’s The Low Code Revolution

John Liu on how Flow Studio MCP lets AI agents read, debug, and build Power Automate flows without opening the portal.

What’s next

• Debug Power Automate Flows with MCP, Step-by-step guide to finding and fixing flow failures
• Build Power Automate Flows with MCP, Create and deploy flows from natural language
• MCP Tools Reference, Complete catalog of all 30+ tools
• Copilot Agent Skills, Pre-built skills for GitHub Copilot