Debashish Ghosal released MCPlex, a stateless HTTP proxy that bridges pre-AI software tools to AI coding agents via the Model Context Protocol (MCP). By mapping simple JSON REST endpoints to MCP tools, the proxy allows developers to use existing command-line interfaces or dashboards within their code editors through AI agents like Claude Code or Cursor. This infrastructure aims to reduce developer friction by eliminating the need to switch contexts between terminal commands, dashboards, and IDEs.

The proxy tool, published as an MIT-licensed package on PyPI under the name `mcplex-backplane`, uses a YAML-based configuration to define connections. This design supports the majority of tools that expose REST endpoints. For tools lacking a server, such as CLI-only applications, the system includes support for native Python connectors or a thin adapter layer consisting of approximately 40 lines of code to expose the necessary JSON interfaces. v0.4.0 of the software introduced safety features including audit logging, rate limiting, and header-based client identity propagation to track which agents initiate calls.

The author identified a significant adoption gap in his previous tools—an incident commander, a CI diagnoser, a code governance checker, and a DORA metrics dashboard—which saw only 20% usage due to interface friction. Despite having functional backends, these tools were designed as independent CLI applications or HTML dashboards rather than API-driven services. To integrate them into the MCP ecosystem, the author implemented standard REST adapters in each repository. The system manages the discovery handshake required by the MCP protocol, allowing agents to list and call tools as if they were natively available features of the development environment.

Development of the project highlighted that implementation of the MCP wire protocol—a JSON-RPC over HTTP interface—requires less effort than managing operational overhead such as service orchestration. The author noted that Docker configuration and documenting environment variables for older services consumed significantly more time than the protocol implementation itself. Future updates for the project include plans for deeper backend integration, better pagination, and support for PUT and DELETE requests, although production-hardened features like OAuth remain unimplemented in the current version.

Debashish Ghosal released MCPlex, a stateless HTTP proxy that bridges pre-AI software tools to AI coding agents via the Model Context Protocol (MCP). By mapping simple JSON REST endpoints to MCP tools, the proxy allows developers to use existing command-line interfaces or dashboards within their code editors through AI agents like Claude Code or Cursor. This infrastructure aims to reduce developer friction by eliminating the need to switch contexts between terminal commands, dashboards, and IDEs.

The proxy tool, published as an MIT-licensed package on PyPI under the name `mcplex-backplane`, uses a YAML-based configuration to define connections. This design supports the majority of tools that expose REST endpoints. For tools lacking a server, such as CLI-only applications, the system includes support for native Python connectors or a thin adapter layer consisting of approximately 40 lines of code to expose the necessary JSON interfaces. v0.4.0 of the software introduced safety features including audit logging, rate limiting, and header-based client identity propagation to track which agents initiate calls.

The author identified a significant adoption gap in his previous tools—an incident commander, a CI diagnoser, a code governance checker, and a DORA metrics dashboard—which saw only 20% usage due to interface friction. Despite having functional backends, these tools were designed as independent CLI applications or HTML dashboards rather than API-driven services. To integrate them into the MCP ecosystem, the author implemented standard REST adapters in each repository. The system manages the discovery handshake required by the MCP protocol, allowing agents to list and call tools as if they were natively available features of the development environment.

Development of the project highlighted that implementation of the MCP wire protocol—a JSON-RPC over HTTP interface—requires less effort than managing operational overhead such as service orchestration. The author noted that Docker configuration and documenting environment variables for older services consumed significantly more time than the protocol implementation itself. Future updates for the project include plans for deeper backend integration, better pagination, and support for PUT and DELETE requests, although production-hardened features like OAuth remain unimplemented in the current version.