We built Flow Studio MCP to give AI agents the same visibility you have
in the Power Automate portal (and a bit more!). The Graph API only returns top-level run status — your agent can't see action inputs, outputs, or nested failures.
Flow Studio MCP exposes all of it.
What your Codex agent can do with it:
- Debug why a flow failed — trace from parent to child flow to the
failing loop iteration
- Inspect action-level inputs and outputs (the key thing the Graph API
doesn't give you)
- Build and deploy new flows from natural language
- Resubmit failed runs, cancel stuck runs
Install in Codex:
1. Configure MCP in ~/.codex/config.toml:
[mcp_servers.flowstudio]
url = "https://mcp.flowstudio.app/mcp"
[mcp_servers.flowstudio.http_headers]
x-api-key = "<YOUR_TOKEN>"
2. Install skills inside a Codex session:
$skill-installer install https://github.com/ninihen1/power-automate-mcp-skills/tree/master/skills/power-automate-mcp
Get your token at https://mcp.flowstudio.app
Real examples from production:
- Expression error in child flow — agent traced nested failure the
portal couldn't surface to an agent
- User reported "bugs" that were data entry errors — agent proved the
flow was correct in seconds
- split() on null crashed 38% of records — agent found it by scanning
all loop iterations
GitHub: https://github.com/ninihen1/power-automate-mcp-skills
Also available on: skills.sh, awesome-copilot, Smithery, ClawHub