Connect an AI Agent

Fluxify has an MCP server. MCP is a standard way for AI apps and coding agents to use outside tools. Once connected, an agent can read your projects, build routes and workflows, run tests and manage members, all as you.

For agents

Read llms.txt first. It lists every docs page, so you can fetch only the ones you need.

The URL

The MCP server lives on your Fluxify address, at /_/admin/mcp:

https://your-domain.com/_/admin/mcp

On a local install that is http://localhost:<port>/_/admin/mcp. Cloud apps such as claude.ai need a public https:// address. See Connecting AI clients.

Connect with OAuth (recommended)

Add the URL to your client. The first time it connects, your browser opens. Sign in to Fluxify and approve the app. That is all.

Claude Code

bash

claude mcp add --transport http fluxify https://your-domain.com/_/admin/mcp

Claude Desktop and claude.ai

Open Settings → Connectors → Add custom connector and paste the URL.

Cursor (~/.cursor/mcp.json)

json

{
  "mcpServers": {
    "fluxify": { "url": "https://your-domain.com/_/admin/mcp" }
  }
}

Connect with an access token

Use a token when the client can't open a browser, such as a script or a CI job.

In Fluxify, open Account → Access tokens and create one. It starts with flx_.
Copy it now. It is shown only once.
Send it as a Bearer header:

bash

claude mcp add --transport http fluxify https://your-domain.com/_/admin/mcp \
  --header "Authorization: Bearer flx_your_token"

json

{
  "mcpServers": {
    "fluxify": {
      "url": "https://your-domain.com/_/admin/mcp",
      "headers": { "Authorization": "Bearer flx_your_token" }
    }
  }
}

Treat the token like a password. Anyone who has it can act as you.

What the agent can do

The agent gets your access, no more. In each project it can do what your role there allows:

Viewer reads routes, workflows, triggers, custom blocks, middlewares, test suites and logs.
Creator also reads app config, integrations, members and packages, changes things and runs tests.
Project Admin also manages members, packages and project settings.

If the agent asks for something your role does not allow, it gets an error like "You need the Creator role in this project." Ask a project admin for the role.

Every tool

"Real" means the tool does something outside Fluxify's own data: it sends a request, runs your code or reaches the internet.

Tool	What it does	Minimum role
`whoami`	Who the agent is signed in as, and your project roles	Any signed-in user
`get_instance_info`	Edition, licence and version	Viewer
`get_block_schemas`	The built-in blocks and their fields	Viewer
`search_docs`	Search the Fluxify docs; returns the matching sections	Viewer
`read_doc`	One docs page's list of sections, or one section's text	Viewer
`get_integration_schema`	The fields one integration type needs	Viewer
`list_projects`	Your projects	Viewer
`get_project`	One project's name and description	Viewer
`get_system_logs`	Compile results and other errors, including failed recorded runs	Viewer
`list_routes`	A project's routes	Viewer
`get_route`	One route's settings and request schemas	Viewer
`list_workflows`	A project's workflows	Viewer
`get_workflow`	One workflow's settings	Viewer
`list_triggers`	A project's triggers	Viewer
`get_trigger`	One trigger's settings	Viewer
`list_custom_blocks`	A project's custom blocks	Viewer
`get_custom_block`	One custom block's inputs and docs	Viewer
`list_middlewares`	A project's middlewares	Viewer
`get_middleware`	One middleware and its blocks	Viewer
`list_test_suites`	The test suites of a route or workflow	Viewer
`get_test_suite`	One test suite	Viewer
`get_canvas`	The blocks and connections of a canvas	Viewer
`list_app_config`	A project's app config keys (no values)	Creator
`get_app_config`	One app config entry (secrets stay masked)	Creator
`list_integrations`	A project's integrations	Creator
`get_integration`	One integration's settings	Creator
`list_members`	A project's members and roles	Creator
`list_packages`	A project's npm packages	Creator
`get_test_runs`	A test suite's recent runs and results, with each case's trace id	Creator
`list_recordings`	A route's or workflow's recorded runs	Creator
`get_recording`	One recorded run, block by block	Creator
`save_route`	Create or change a route	Creator
`delete_route`	Delete a route	Creator
`save_workflow`	Create or change a workflow	Creator
`delete_workflow`	Delete a workflow	Creator
`save_trigger`	Create or change a trigger	Creator
`delete_trigger`	Delete a trigger	Creator
`save_middleware`	Create or change a middleware	Creator
`delete_middleware`	Delete a middleware	Creator
`save_custom_block`	Create or change a custom block's details	Creator
`delete_custom_block`	Delete a custom block	Creator
`save_app_config`	Create or change an app config entry	Creator
`delete_app_config`	Delete an app config entry	Creator
`save_integration`	Create or change an integration	Creator
`delete_integration`	Delete an integration	Creator
`test_integration_connection`	Check an integration connects (real)	Creator
`get_integration_schema_details`	Tables, columns and keys of a database integration (no rows)	Creator
`kv_get`	Read one key from a KV integration, with its expiry	Creator
`edit_canvas`	Add, change and connect blocks on a canvas	Creator
`call_route`	Send a request to a route (real)	Creator
`run_test_suite`	Run a test suite (real)	Creator
`add_member`	Add a user to a project	Project Admin
`update_member_role`	Change a member's role	Project Admin
`remove_member`	Remove a member	Project Admin
`install_package`	Install npm packages (real)	Project Admin
`remove_package`	Uninstall npm packages	Project Admin
`update_project`	Change a project's name, description or hidden flag	Project Admin

Safety

Runs are real

call_route and run_test_suite run your code for real. If a route writes to a database, sends an email or calls another API, that happens. Point agents at a test project when you can.

Debug errors

When a route fails, its callers get a short message such as failed to execute native db block, never the database error or a stack trace. call_route also returns the real error: the block that failed, its message, the cause behind it (such as the SQL error), and, for an error in your own code, where in that code it happened. Only a creator calling through the tools gets it; anyone calling the route's URL, whatever headers they send, still gets the short message.

You approve every app. An app can't connect until you sign in and say yes in your browser.
Your client may ask before each change. Tools that delete or run things are marked as such, so most clients ask you first.
Revoke access any time. Open Account → Connected apps to remove an app, or Account → Access tokens to delete a token. It stops working at once.
Apps stay signed in for up to 30 days. A connected app gets a short-lived token and renews it by itself. Each renewal hands out a new one and retires the old one. If the app goes unused for 30 days, you sign in again. Removing the app ends all of this at once.

Connect an AI Agent ​

The URL ​

Connect with OAuth (recommended) ​

Connect with an access token ​

What the agent can do ​

Every tool ​

Safety ​