Can I deploy from Cursor or Claude?

Yes. Connect dFlow in Cursor, Claude, Copilot, Codex, Antigravity, or a similar editor. Ask it to inspect a deploy, review logs, find the root cause, and ship the next change without leaving chat.

Which AI editors work with dFlow MCP?

Cursor, Claude Code, GitHub Copilot, Codex, Antigravity, and any editor that can connect this way. Connect once. You get the same tools as the dashboard in each of them.

Do I need to write Dockerfiles or YAML?

No. Ask your editor to deploy the repo. We find the stack, build it, and keep it running. You review the change before it ships.

Can I debug production from the editor?

Yes. Read deploy logs and runtime logs in chat, fix the code, and deploy again from the same conversation. You can review the change before it goes out.

What infrastructure does the agent deploy to?

The machine bound to the environment you pick — dFlow servers, your cloud, or a machine you already have. The editor uses your role. Calls go to that environment.

Do I still need a DevOps team?

Not for day-to-day shipping. Connect your editor and ask it to deploy. Some server know-how helps, but nobody needs infrastructure as a full-time job.

mcp · oauth/any MCP client

AI workflow

Ship from the editor you already use.

Type in Cursor or Claude. Your editor checks the deploy, reads the logs, and ships the next change.

Connect your editor Read MCP setup

Agent

Platform ops chat for this organisation — apps, environments, services, deploys, and logs.

Agent connectedAsk about apps, environments, services, deploys, or logs.

dFlow Agent

Ask about apps, environments, services, deploys, or logs. Try a suggestion below to get started.

What I can do

Read-only ops over this organisation.

Platform lookups

List applications and templates
Inspect environments, services, and deployments
Fetch deploy and runtime logs, then diagnose failures

List my applicationsList templatesList environments for my latest app

Ask about your dFlow stack. Use / or @ for tools.

New chatSend

In the dashboard

Ask about the running app.

See the apps, the deploys, and the logs without leaving the product. The editor uses the same tools after you connect it.

idledemo · no network

Connected over OAuth. Send a prompt. The agent calls the same tools as the dashboard. Nothing here hits the network.

mcp connected

Send a prompt. tools/call params land here.

envprod

nodeworker-prod

clientcursor

authoauth

t+1 session

Recorded session

Ask once. Production is running.

The editor inspects the deploy, reviews the logs, and ships the next change. You keep writing.

idle

create_app_serviceid 01

params

namestorefront-web

environmentIdenv_prod

result

serviceIdsvc_8f2a

typeapp

statecreated

t+2 wire

Client config

Connect the editor. Keep writing.

Drop this into Cursor, Claude Code, Copilot, Codex, or Antigravity. Authorize once. After that, you ship from chat.

CursorClaude CodeGitHub CopilotCodexAntigravityany MCP client

streamable-http

{
  "mcpServers": {
    "dflow": {
      "url": "https://app.dflow.sh/api/mcp"
    }
  }
}

Opens the app

CursorOpen CopilotOpen

Copies the command

t+3 methods

Method directory

The same tools as the dashboard.

Filter a method, read what it does. Names match the server. Nothing here is invented.

69/69

Applications6

Environments5

Worker nodes3

Services23

Deployments6

Git providers6

Templates5

Docker registries3

Backups12

inspect

method

create_app_service

Create app service

Apps, databases, Docker images, env, domains, scale. The daily surface.

use

Spin up an app service on the environment’s worker node, then set git or file source.

{
  "method": "tools/call",
  "params": {
    "name": "create_app_service"
  }
}

t+4 scope

Session contract

Your editor. Your role. That machine.

One consent. The same access you already have. Calls land on the environment you picked.

01 · consent
One OAuth screen
The editor asks once. You can revoke the client from the dashboard. No long-lived personal tokens sitting in a dotfile.
02 · grants
The agent inherits your role
It gets the same create, read, update, and delete you have. Nothing above it. Encrypted variables are redacted on read.
03 · compute
Calls land on that environment’s node
Every tool runs against the worker node bound to the environment you picked in the workspace.

t+5 recover

The new DevOps loop

Fix production from the editor.

When something breaks, the logs are already in chat. Fix it. Ship again. You do not leave the work.

crash

14:02:11get_service_runtime_logs

storefront-web · tail 42 · worker-node i-9f2

ERROR ECONNREFUSED 127.0.0.1:5432 src/db.ts:18

GET /checkout 500 18ms

01
Logs streamed to your editor
02
Diff-friendly fixes you can review
03
Ship the patch back to prod in one flow

t+6 notes

FAQ

AI workflow questions

Editors, what you can do from chat, and where the app runs.

t+7 continue

live

Connect your editor and ship

Open Cursor, Claude, Copilot, or Codex. Connect dFlow. Ask it to inspect, debug, and deploy.

Connect your editor Read MCP setup

mcp · oauth/any MCP client

AI workflow

Ship from the editor you already use.

Type in Cursor or Claude. Your editor checks the deploy, reads the logs, and ships the next change.

Connect your editor Read MCP setup

Agent

Platform ops chat for this organisation — apps, environments, services, deploys, and logs.

Agent connectedAsk about apps, environments, services, deploys, or logs.

dFlow Agent

Ask about apps, environments, services, deploys, or logs. Try a suggestion below to get started.

What I can do

Read-only ops over this organisation.

Platform lookups

List applications and templates
Inspect environments, services, and deployments
Fetch deploy and runtime logs, then diagnose failures

List my applicationsList templatesList environments for my latest app

Ask about your dFlow stack. Use / or @ for tools.

New chatSend

In the dashboard

Ask about the running app.

See the apps, the deploys, and the logs without leaving the product. The editor uses the same tools after you connect it.

idledemo · no network

Connected over OAuth. Send a prompt. The agent calls the same tools as the dashboard. Nothing here hits the network.

mcp connected

Send a prompt. tools/call params land here.

envprod

nodeworker-prod

clientcursor

authoauth

t+1 session

Recorded session

Ask once. Production is running.

The editor inspects the deploy, reviews the logs, and ships the next change. You keep writing.

idle

create_app_serviceid 01

params

namestorefront-web

environmentIdenv_prod

result

serviceIdsvc_8f2a

typeapp

statecreated

t+2 wire

Client config

Connect the editor. Keep writing.

Drop this into Cursor, Claude Code, Copilot, Codex, or Antigravity. Authorize once. After that, you ship from chat.

CursorClaude CodeGitHub CopilotCodexAntigravityany MCP client

streamable-http

{
  "mcpServers": {
    "dflow": {
      "url": "https://app.dflow.sh/api/mcp"
    }
  }
}

Opens the app

CursorOpen CopilotOpen

Copies the command

t+3 methods

Method directory

The same tools as the dashboard.

Filter a method, read what it does. Names match the server. Nothing here is invented.

69/69

Applications6

Environments5

Worker nodes3

Services23

Deployments6

Git providers6

Templates5

Docker registries3

Backups12

inspect

method

create_app_service

Create app service

Apps, databases, Docker images, env, domains, scale. The daily surface.

use

Spin up an app service on the environment’s worker node, then set git or file source.

{
  "method": "tools/call",
  "params": {
    "name": "create_app_service"
  }
}

t+4 scope

Session contract

Your editor. Your role. That machine.

One consent. The same access you already have. Calls land on the environment you picked.

01 · consent
One OAuth screen
The editor asks once. You can revoke the client from the dashboard. No long-lived personal tokens sitting in a dotfile.
02 · grants
The agent inherits your role
It gets the same create, read, update, and delete you have. Nothing above it. Encrypted variables are redacted on read.
03 · compute
Calls land on that environment’s node
Every tool runs against the worker node bound to the environment you picked in the workspace.

t+5 recover

The new DevOps loop

Fix production from the editor.

When something breaks, the logs are already in chat. Fix it. Ship again. You do not leave the work.

crash

14:02:11get_service_runtime_logs

storefront-web · tail 42 · worker-node i-9f2

ERROR ECONNREFUSED 127.0.0.1:5432 src/db.ts:18

GET /checkout 500 18ms

01
Logs streamed to your editor
02
Diff-friendly fixes you can review
03
Ship the patch back to prod in one flow

t+6 notes

FAQ

AI workflow questions

Editors, what you can do from chat, and where the app runs.

t+7 continue

live

Connect your editor and ship

Open Cursor, Claude, Copilot, or Codex. Connect dFlow. Ask it to inspect, debug, and deploy.

Connect your editor Read MCP setup

Ship from the editor you already use.

Ask about the running app.

Ask once. Production is running.

Connect the editor. Keep writing.

The same tools as the dashboard.

create_app_service

Your editor. Your role. That machine.

One OAuth screen

The agent inherits your role

Calls land on that environment’s node

Fix production from the editor.

AI workflow questions

Can I deploy from Cursor or Claude?

Which AI editors work with dFlow MCP?

Do I need to write Dockerfiles or YAML?

Can I debug production from the editor?

What infrastructure does the agent deploy to?

Do I still need a DevOps team?

Connect your editor and ship

Ship from the editor you already use.

Ask about the running app.

Ask once. Production is running.

Connect the editor. Keep writing.

The same tools as the dashboard.

create_app_service

Your editor. Your role. That machine.

One OAuth screen

The agent inherits your role

Calls land on that environment’s node

Fix production from the editor.

AI workflow questions

Can I deploy from Cursor or Claude?

Which AI editors work with dFlow MCP?

Do I need to write Dockerfiles or YAML?

Can I debug production from the editor?

What infrastructure does the agent deploy to?

Do I still need a DevOps team?

Connect your editor and ship