New

Just launched: Vueblocks, a brand-new search engine for Vue packages! Explore Vueblocks →

particle-academy/fancy-flow-mcp

A Laravel MCP server that lets an AI agent build fancy-flow workflows headlessly — list node kinds, add/connect/configure nodes, validate, export the WorkflowSchema, and run — as a thin, deterministic wrapper over particle-academy/fancy-flow-php.
1,226
Install
composer require particle-academy/fancy-flow-mcp
Latest Version:v0.6.0
PHP:^8.4
Laravel:12 · 13
License:MIT
Last Updated:Nov 19, 2026
Links: GitHub  ·  Packagist
Maintainer: wishborn

fancy-flow-mcp

Fancified

A Laravel MCP server that lets an AI agent build fancy-flow workflows headlessly.

Built on laravel/mcp (Laravel's official MCP package) and powered by particle-academy/fancy-flow-php (the PHP runtime twin of fancy-flow). The connecting model is the only intelligence involved — every tool here is a thin, deterministic wrapper over fancy-flow-php's schema, registries, validation, and engine. No LLM calls, no network, no guessing.

An agent discovers the node kinds, creates a workflow, adds and wires nodes (connections are port-validated), fills in each node's config (schema-checked), validates the whole graph, and exports a portable WorkflowSchema v1 JSON — the same document <FlowEditor> imports and both the Node and PHP engines run unchanged.

composer require particle-academy/fancy-flow-mcp

Requires PHP 8.4+ (fancy-flow-php's floor) and Laravel 13+.


For agents. This package is listed on the Fancy MCP catalog server at https://ui.particle.academy/mcp (streamable HTTP, no API key; MCP id fancy-ui), with install commands and docs, alongside the rest of the kit. How an agent installs and uses this package is in AGENTS.md.

The tools

Tool What it does
list_node_kinds Every authorable kind — id, category, label, ports, aliases. Optional category filter.
describe_node_kind One kind's full config schema, ports, defaults, and aliases.
create_workflow Start an empty draft; returns a workflow_id used by every later call.
import_workflow Load an existing WorkflowSchema document into an editable draft (lenient — reports issues, doesn't reject).
get_workflow Inspect the current draft — metadata, nodes, edges.
list_workflows Every draft the server holds.
delete_workflow Discard a draft.
add_node Add a node of a kind; config defaults applied, config schema-checked.
configure_node Set/merge a node's config, re-validated against its kind.
connect_nodes Draw a port-validated edge (handles default out → in).
remove_node Delete a node and cascade its edges.
remove_edge Delete a single edge.
validate_workflow Authoritative Workflow::import check (unknown kinds, missing config, dangling edges) + host capability status.
export_workflow The deliverable: canonical WorkflowSchema v1 JSON.
run_workflow Smoke-test the graph with fancy-flow-php's deterministic fake executors.

Plus a resource — flow://node-kinds — carrying the entire kind catalogue as one JSON document, so an agent can pull it once as context.

Why "port-validated" connections matter

Workflow::import() validates that an edge's endpoints exist; it does not check that a handle is a real port. connect_nodes adds that: a branch needs source_handle "true" or "false", a merge needs target_handle "a" or "b", a terminal output can't be a source, a trigger can't be a target — and when you get it wrong, the error names the valid ports. Dynamic-port kinds are resolved from config the way the engine documents them (llm_router routes, switch_case cases, subflow stream mode).

Registering the server

The service provider registers a local (stdio) server named fancy-flow out of the box, so it works with a desktop AI client immediately:

php artisan mcp:inspector fancy-flow      # try it interactively
php artisan mcp:start fancy-flow          # (clients usually start this for you)

Publish the config to customize:

php artisan vendor:publish --tag=fancy-flow-mcp-config
// config/fancy-flow-mcp.php
'local' => ['enabled' => true, 'name' => 'fancy-flow'],
'web'   => ['path' => null, 'middleware' => []], // set a path to expose over HTTP
'store' => ['driver' => 'cache', /* 'array' for a long-lived local server */ ],

To expose it over HTTP instead (protect it — an MCP endpoint runs tools on your app), set web.path (e.g. /mcp/fancy-flow) and appropriate middleware, or register it yourself in routes/ai.php:

use FancyFlow\Mcp\Server\FlowBuilderServer;
use Laravel\Mcp\Facades\Mcp;

Mcp::web('/mcp/fancy-flow', FlowBuilderServer::class)->middleware(['auth:sanctum']);

The draft store

An agent authors a workflow across many tool calls, so drafts persist between them via a DraftStore. cache (default) survives the request-per-call lifecycle of a web server; array keeps drafts in-process, which is fine for a long-lived local/stdio server. Rebind FancyFlow\Mcp\Store\DraftStore in your app to persist to a database instead.

A typical tool-by-tool build, for an agent connected to the server, is in AGENTS.md.

License

MIT © Particle Academy

Related Packages

particle-academy/fancy-flow-php

PHP runtime for fancy-flow workflow graphs — the framework-free twin of @particl...

4,080 1
padosoft/askmydocs-mcp-pack

Framework-agnostic MCP (Model Context Protocol) plumbing for Laravel — contracts...

8,590 0
contributte/mcp

Model Context Protocol (MCP) server integration for Nette Framework

3,485 2
onelearningcommunity/laravel-model-explorer

A zero-config browser UI and MCP server for exploring your Eloquent models, rela...

122,093 7
mattiasgeniar/filament-mcp

Expose your Filament resources to AI agents over the Model Context Protocol, wit...

8,482 13

Version History

Version Released PHP License
v0.6.0 ^8.4 MIT
v0.5.0 ^8.4 MIT
v0.4.0 ^8.4 MIT
v0.3.1 ^8.4 MIT
v0.3.0 ^8.4 MIT