Universal MCP Toolkit

What Markgatcha/universal-mcp-toolkit MCP server does

Markgatcha/universal-mcp-toolkit MCP server is an open-source TypeScript monorepo and command-line toolkit for working with multiple MCP servers from one setup. It includes a shared runtime, a server registry, configuration generation, execution commands, discovery features, and composition utilities. The repository lists integrations spanning collaboration, productivity, data, platform, research, local-file, and memory use cases, including GitHub, Slack, Notion, databases, cloud platforms, and FileSystem.

The toolkit is intended for developers who need a consistent way to select MCP servers, produce host configuration, run servers locally, and connect tools across server boundaries. It is also presented as a reference implementation with shared validation, logging, error handling, and packaging conventions.

How it works

The universal-mcp-toolkit CLI provides commands for listing available servers, interactively installing a selection, generating configuration for Claude Desktop, running a selected server, checking its environment, and validating workflows. Tool discovery can search across the available servers by name or category. Registry-compatible and local discovery modes are also supported.

Cross-server workflows are represented as versioned JSON files with validated inputs and step references. The toolkit also retains a simpler composition mode for piping output between two steps. Its session abstraction can coordinate calls to several MCP servers in parallel, while typed tool chains support compile-time-checked pipelines.

At the runtime level, the toolkit supports stdio and HTTP+SSE transport handling. It includes lazy loading for server packages and tool registration, TTL/LRU result caching, reconnect and circuit-breaker behavior for transport failures, and structured bridge errors containing tool and argument context. Streaming calls expose completed result chunks through an asynchronous iterator; the README distinguishes this from protocol-level streaming.

Setup and configuration

The quick-start examples use the npm package through npx, with commands such as npx universal-mcp-toolkit list, install, config, run, doctor, and workflow validate. A global npm installation is also documented, after which the executable is available as umt.

Interactive setup can select servers, choose a transport, and write configuration. The config command can generate a Claude Desktop configuration snippet for named servers. Individual integrations may require their own credentials; the README’s server table identifies examples such as GITHUB_TOKEN, NOTION_TOKEN, SLACK_BOT_TOKEN, and LINEAR_API_KEY, but the required variables depend on the selected server.

Tools and capabilities

• Discover servers and tools through the CLI
• Generate host configuration for selected MCP servers
• Run servers over stdio or HTTP+SSE
• Validate and run JSON-defined multi-server workflows
• Compose short tool pipelines
• Coordinate calls through multi-server sessions
• Cache tool results with TTL and LRU behavior
• Diagnose server environments and transport problems
• Load server packages and expensive tools lazily

Limitations and notes

The toolkit is an aggregator and runtime layer, not a replacement for the credentials or external services used by its individual integrations. Configuration requirements therefore vary by server. Its documented chunk iterator does not represent MCP protocol-level streaming, and ordinary tool-validation errors are not treated as transport failures by the circuit breaker.