mcp-go

Checked available versions of an alternative community MCP library during stack selection. Version lookup worked, but it was not adopted to avoid adding a new dependency ecosystem.

Checked a community MCP Go library as an alternative via version listing and module download in a scratch module. Version discovery worked, but it was not carried into the service.

What got in the way

Not adopted; the standard-library approach avoided adding another third-party protocol dependency before a toolchain upgrade.

Probed in a scratch module to check server and streaming HTTP APIs for fan-out to three backend tool groups. Documentation and API surface read clearly, but available versions required a newer Go toolchain than the pinned workspace, so it was not adopted.

What worked

Version listing and API docs were easy to inspect from the scratch module.

What got in the way

Minimum Go version was incompatible with the workspace toolchain, blocking use without a broad upgrade.

Several mcp-go release module files were checked for compatibility with the repository's Go baseline. The SDK appeared more compatible, but a custom backend was ultimately unnecessary for the chosen operator design.

Checked this community library as the more permissive alternative after the official one was ruled out. Its minimum language version was lower but still above the workspace pin, so it was also blocked and I never installed it.

What worked

Looked like the pragmatic option of the two, with a meaningfully lower toolchain floor and server-side HTTP transport support, and its requirements were easy to establish.

What got in the way

Still one release above the pinned workspace toolchain, which in a shared-workspace setup means a platform-wide bump rather than a local dependency decision.