FlowDot

What the FlowDot MCP server does

FlowDot MCP server exposes FlowDot’s workflow platform to AI clients that support the Model Context Protocol. Agents can find workflows, inspect their details and input schemas, create new workflows, modify graph structure, validate configurations, and execute workflows with optional inputs.

The available operations also cover execution oversight and workflow collaboration. An agent can retrieve execution results and history, stream live execution events over SSE, cancel pending or active runs, retry failed runs, and review workflow metrics, comments, ratings, tags, favorites, and public visibility. Search tools cover workflows and other FlowDot resources such as apps and custom nodes.

The README describes a broader platform interface with 150 tools in 17 categories. In addition to workflows, those categories include custom nodes, React-based apps, agent recipes, knowledge bases, agent toolkits, agent characters for voice calls, and community sharing. The tool list supplied for this listing focuses on workflow and node operations.

How it works

An MCP client starts the server locally and communicates with it through the configured command. The server authenticates requests with a FlowDot token, then exposes FlowDot actions as callable MCP tools. Clients can first inspect available node types and individual node schemas, create or retrieve a workflow graph, add or update nodes, connect node inputs and outputs, and run validation before execution.

Execution can be synchronous or asynchronous. Asynchronous runs can be checked with status and history tools, followed through real-time SSE streaming, cancelled, or retried with the original inputs. Workflow discovery can use names, descriptions, tags, public listings, or unified search.

The server also provides eight educational learn:// resources covering FlowDot concepts such as workflows, recipes, custom nodes, apps, toolkits, knowledge bases, and agent characters. These resources can be read through the MCP resource interface before calling tools.

Setup and configuration

Create a FlowDot account and generate an MCP token from the MCP Tokens settings. The token begins with fd_mcp_; scopes can be selected when the token is created. Provide it to the client as the FLOWDOT_API_TOKEN environment variable.

For Claude Code, the documented setup uses claude mcp add with npx -y @flowdot.ai/mcp-server. Other clients can launch the package with npx, install it globally with npm, or use the Claude Desktop .mcpb extension bundle. After changing configuration, restart the client so it loads the server. Claude Code uses the configuration on its next session.

The free plan is documented as including five workflow runs per day and ten toolkit calls per day. The README also lists a Creator plan with monthly and daily limits; current plan details should be checked on FlowDot’s site.

Tools and capabilities

The FlowDot MCP server includes tools for:

• Listing, searching, inspecting, duplicating, tagging, favoriting, publishing, and deleting workflows.
• Creating workflow graphs and adding, updating, deleting, and connecting nodes.
• Reading node types and full node schemas, including inputs, outputs, and properties.
• Validating workflows for disconnected nodes, missing connections, and invalid configuration.
• Executing, streaming, monitoring, cancelling, and retrying workflow runs.
• Reviewing execution history, metrics, comments, and ratings.
• Managing custom nodes, including creation, script updates, visibility, community discovery, and comments.
• Creating and editing FlowDot React apps and linking them to workflows or toolkits, as described in the README.

Limitations and notes

Recipes can be designed through MCP but are not executed by the MCP server according to the README. Their execution requires the FlowDot CLI because recipes are long-running agentic programs that may exceed AI client timeouts and need local file, code, or shell access.

Destructive operations such as deleting workflows, nodes, custom nodes, or apps should be used carefully. The supplied material does not document a local workflow file format, offline mode, or client-specific permission behavior. Token scopes are selectable during token creation, so the available operations depend on the permissions granted to the token.