FlowStudio MCP — Power Automate Skills for AI Agents

Give your AI agent the same visibility you have in the Power Automate portal — plus a bit more. The Graph API only returns top-level run status — agents can't see action inputs, loop iterations, or nested failures. FlowStudio MCP exposes all of it.

You can click through the portal and find the root cause. Your agent can't — unless it has MCP.

When you need this

• Your agent can see that a flow failed, but not why — Graph API only returns status codes
• You want your agent to see action-level inputs and outputs, like you can in the portal
• A loop has hundreds of iterations and some produced bad output — in the portal you'd click through each one, but the agent can scan all iteration inputs and outputs at once
• You want to check flow health, failure rates, and maker activity across your tenant without opening the admin center
• You need to classify flows, detect orphaned resources, or audit connectors at scale — without installing the CoE Starter Kit
• You're tired of being the middle-man between your agent and the portal

Graph API vs FlowStudio MCP

The core difference: Graph API gives your agent run status. MCP gives your agent the inputs and outputs of every action.

What the agent sees	Graph API	FlowStudio MCP
Run passed or failed	Yes	Yes
Action inputs and outputs	No	Yes
Error details beyond status code	No	Yes
Child flow run details	No	Yes
Loop iteration data	No	Yes
Flow definition (read + write)	Limited	Full JSON
Resubmit / cancel runs	Limited	Yes
Cached flow health & failure rates	No	Yes
Maker / Power Apps / connection inventory	No	Yes
Governance metadata (tags, impact, owner)	No	Yes

Skills

Skill	Description
power-automate-mcp	Connect to and operate Power Automate cloud flows — list flows, read definitions, check runs, resubmit, cancel
power-automate-debug	Step-by-step diagnostic process for investigating failing flows
power-automate-build	Build, scaffold, and deploy Power Automate flow definitions from scratch
power-automate-monitoring	Flow health, failure rates, maker inventory, Power Apps, environment and connection counts
power-automate-governance	Classify flows by impact, detect orphans, audit connectors, manage notifications, compute archive scores

The first three skills use live Power Automate API calls. The monitoring and governance skills use the cached store — a daily snapshot with aggregated stats, remediation hints, and governance metadata. Requires a FlowStudio for Teams or MCP Pro+ subscription for store tools.

Each skill follows the Agent Skills specification and works with any compatible agent.

Supported agents

Copilot, Claude Code, ChatGPT, Codex, OpenClaw, Gemini CLI, Cursor, Goose, Amp, OpenHands

Guides

Full walkthroughs on learn.flowstudio.app:

• Getting started — install FlowStudio MCP for any agent
• Build flows with AI agents — a worked 3-step example
• Vibe code Power Automate — describe the flow, the agent builds it
• Power Automate in VS Code — work on flows from your editor
• Debug a failing flow — find the root cause from action-level data
• By agent: ChatGPT · Claude · GitHub Copilot · Copilot Studio · Codex

Quick Start

Connect from ChatGPT or claude.ai

Sign in with your Microsoft account instead of pasting a key. Add https://mcp.flowstudio.app/mcp/oauth as a connector in ChatGPT (Developer mode on) or at claude.ai, then follow the sign-in prompt. Walkthrough: ChatGPT for Power Automate.

Install as Claude Code plugin

Available through the Claude plugin marketplace after approval. To test locally:

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_TOKEN>"

Get your token at mcp.flowstudio.app.

Install in Codex

Inside a Codex session, install skills directly:

$skill-installer install https://github.com/ninihen1/power-automate-mcp-skills/tree/main/skills/power-automate-mcp
$skill-installer install https://github.com/ninihen1/power-automate-mcp-skills/tree/main/skills/power-automate-debug
$skill-installer install https://github.com/ninihen1/power-automate-mcp-skills/tree/main/skills/power-automate-build
$skill-installer install https://github.com/ninihen1/power-automate-mcp-skills/tree/main/skills/power-automate-monitoring
$skill-installer install https://github.com/ninihen1/power-automate-mcp-skills/tree/main/skills/power-automate-governance

Then connect the MCP server in ~/.codex/config.toml:

[mcp_servers.flowstudio]
url = "https://mcp.flowstudio.app/mcp"

[mcp_servers.flowstudio.http_headers]
x-api-key = "<YOUR_TOKEN>"

Install via skills.sh

Search for flowstudio on skills.sh, or:

npx skills add github/awesome-copilot -s flowstudio-power-automate-mcp
npx skills add github/awesome-copilot -s flowstudio-power-automate-debug
npx skills add github/awesome-copilot -s flowstudio-power-automate-build
npx skills add github/awesome-copilot -s flowstudio-power-automate-monitoring
npx skills add github/awesome-copilot -s flowstudio-power-automate-governance

Install via ClawHub

npx clawhub@latest install power-automate-mcp

Install via Smithery

npx smithery skill add flowstudio/power-automate-mcp

Manual install

Copy the skill folder(s) into your project's .github/skills/ directory (or wherever your agent discovers skills).

Connect the MCP server

Claude Code:

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

Codex (~/.codex/config.toml):

[mcp_servers.flowstudio]
url = "https://mcp.flowstudio.app/mcp"

[mcp_servers.flowstudio.http_headers]
x-api-key = "<YOUR_TOKEN>"

Copilot / VS Code (.vscode/mcp.json):

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

Get your token at mcp.flowstudio.app.

Real debugging examples

These are from real production investigations, not demos.

• Expression error in child flow — contains(string(...)) crashed on a nested property. Agent traced through parent flow, into child, through loop iterations, and found the failing input. Portal showed "ExpressionEvaluationFailed" with no context.

• Data entry, not a flow bug — User reported two "bugs" back to back. Agent proved both were data entry errors (missing comma in email, single address in CC field). Flow was correct. Diagnosed in seconds.

• Null value crashes child flow — split(Name, ', ') crashed when 38% of records had null Names. Agent traced parent to child to loop to action, found the root cause, and deployed a fix via update_live_flow.

Prerequisites

• A FlowStudio MCP subscription (all live tools)
• For store tools (monitoring, governance): FlowStudio for Teams or MCP Pro+
• MCP endpoint: https://mcp.flowstudio.app/mcp
• API key / JWT token (passed as x-api-key header)

Repository structure

skills/
  power-automate-mcp/          core connection & operation skill
  power-automate-debug/        debug workflow skill
  power-automate-build/        build & deploy skill
  power-automate-monitoring/   flow health & tenant inventory skill
  power-automate-governance/   compliance & governance skill
examples/                      real debugging walkthroughs
README.md
LICENSE                        MIT

Available on GitHub

Works with Copilot, Claude, ChatGPT, and any MCP-compatible agent.

• awesome-copilot (merged)
• skills.sh (3K+ installs)
• Smithery (published)
• ClawHub (v1.1.0)

Contributing

Contributions welcome. Each skill folder must contain a SKILL.md with the required frontmatter. See the existing skills for the format.

License

MIT

---

Keywords: Power Automate debugging, flow run history, expression evaluation failed, child flow failure, nested action errors, loop iteration output, agent automation MCP, Power Platform AI, flow definition deploy, resubmit failed run, flow monitoring, governance, CoE, orphan detection, connector audit, archive score, maker inventory