dFlow MCP

Written By Charan

Last updated About 4 hours ago

The dFlow MCP server gives AI assistants and compatible IDEs a standard way to work with your dFlow organisation. Actions use the same permissions as the web app—you are not opening a separate "admin API."

Note: New tools appear in product releases. The tables below reflect what dFlow exposes today; your client's tool list should match the current product.

What is MCP?

Model Context Protocol (MCP) is an open standard: AI apps connect to tools (named operations) over HTTP in a consistent way, instead of each product inventing its own integration.

dFlow MCP server

The dFlow MCP server runs on the dFlow dashboard. It is authenticated and tenant-scoped: after OAuth, you work in one organisation at a time.

Base server URL (dFlow Cloud): https://app.dflow.sh/api/mcp

Authentication

dFlow uses OAuth 2.0 only. API keys are not used for MCP.

  1. Sign in to dFlow and open the MCP authorisation flow in the dashboard (wording may vary by version).

  2. Choose your organisation and approve access for your client.

  3. In your MCP client (for example Cursor), complete OAuth and any connection prompts so the client is authorised—typically the client stores what it needs; you do not paste a secret by hand.

Available tools

Templates

Tool

Description

list_templates

List official, community, or personal templates (by type).

get_template_by_id

Fetch a template by id and type.

update_template_by_id

Update a personal template.

create_template

Create a new personal template for the organisation.

Docker registries

Tool

Description

list_docker_registries

List registry integrations (for example Docker Hub, GHCR, DigitalOcean, Quay—whatever your tenant has configured).

create_docker_registry

Add a registry; a connection test runs first. If the test fails, nothing is saved.

update_docker_registry_by_id

Update a registry; same test-before-write behaviour as create.

GitHub and git providers

Tool

Description

list_github_git_providers

List GitHub-type providers for the organisation. Output may be sensitive—handle it like the dashboard.

prepare_github_app_registration

Get manifest, state, and a new GitHub app link to continue in the browser.

get_github_app_install_url

Get an install URL when the app is ready.

list_github_repositories

List repositories for a git provider id.

list_github_branches

List branches for owner / repo for a provider id.

Some GitHub steps still finish in a browser (registration or install) even when MCP returns URLs and state.

Responses usually include human-readable content and, when useful, structuredContent (parsed data) for the client.

Wire protocol

Point your client at the Base server URL using MCP over HTTP (for example Streamable HTTP), as your client's documentation describes. dFlow Cloud uses the URL in this page; self-hosted uses the same path on your dashboard.

Setup instructions

Most clients (for example Cursor)

Add the server URL. The label (here dflow) is only for you:

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

Use ~/.cursor/mcp.json or .cursor/mcp.json depending on how you manage Cursor. After saving, open the client's MCP settings and finish OAuth or connection prompts if asked.

dFlow MCP is OAuth only in the product: we do not support connecting by pasting a key or by hand-editing an Authorization header in config. Your client must complete OAuth with dFlow.

First connection

  1. Finish OAuth in dFlow and the client.

  2. Run a simple read or list tool first, then other tools. If a call fails, see Roles and permissions and try reconnecting the client or repeating sign-in.

Video walkthrough

Short demo: enable the dFlow MCP in Cursor, then use it to create a personal template.

Security and good practice

  • Same roles as the app: if your role cannot do something in the dashboard, expect the matching tool to fail.

  • One org per sign-in; do not mix tenants.

  • Do not paste raw tool output from git or registry tools into public places if it may include secrets.

  • If the product revokes or rotates access, sign in to MCP again in the client.

For contributors

If you ship new MCP tools in the dFlow codebase, register them with the dashboard's MCP integration and keep user-facing names in sync with this page. This doc stays product-focused; use the repository and team practices for implementation detail.

Related

The open MCP standard is described on modelcontextprotocol.io. The sections above are specific to dFlow's dashboard at /api/mcp.