@workglow/mcp
Model Context Protocol tasks and plumbing for Workglow.
Links
README
From the repo.
WORKGLOW
Overview
Simple library to build linear and graph based workflows. It is designed to be simple to use and easy to extend, yet powerful enough to handle task rate limits, retries, etc.
Authors
Roadmap
- Release 0.1.0: Stable API for task(input,config) graph, storage, and job queue.* 31 March 2026
- Release 0.2.0: Stable API for task(config), stable text generation; alpha: entitlements, browser control, toolcalling, and chat loops.* 9 April 2026
- Release 0.3.0: Stable image generation, and job queue refactors. 20 May 2026
- Release 0.x.0: Beta: chat loops, entitlments, user in the loop.
- Release 0.x.0: Stable toolcalling, chat loops, entitlments, user in the loop; beta: agent loops.
- Release 0.x.0: Stable agents.
- Release 0.x.0: Stable Browser control (direct and mcp).
- Release 0.x.0: Reworked output cache, checkpointing, and resumption.
*stable is a relative term, and will be updated at some point as we go before version 1.0.0.
Docs
Technical Deep Dives
- Task Graph DAG Engine
- Dual Mode Execution
- Control Flow Tasks
- Dataflow and Streaming
- Job Queue System
- Composite Rate Limiting
- Storage Abstraction
- Knowledge Base and RAG
- Schema and Input Resolution
- AI Task Framework
- AI Provider System
- Model Registry
- Worker System
- Entitlements System
- Credential Management
- Service Registry and DI
- Event System
- Multi-Runtime Abstraction
- Build System
- Task Registry
- Graph Data Structures
- MCP Integration
Packages
Packages:
-
providers/* — vendor provider packages (Anthropic, OpenAI, Gemini, Ollama, HuggingFace, llama.cpp, MediaPipe, Chrome AI, Cactus (needle-rs), Playwright, Electron, SQLite, Postgres, Supabase, bun-webview)
Examples
CLI
examples/cli is a full CLI built on the library. It runs any saved workflow, agent, or single task as a task graph, and renders that graph live — a row per task with its status, detail, progress and duration, over a bar for the run as a whole.

The same commands are also available in a local web console, served by the CLI itself
with workglow web. It reads the command tree off the live program, so every command
— including those a downstream CLI adds — gets a form built from the same schemas the
terminal prompts from, and the run it starts streams into the same graph view.

Web

Node Editor
Collected info
- Language: TypeScript
- Source updated: 9/1/2026
Config for your environment
Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "{MCP_ENDPOINT_URL}"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.
