MCP is perceived to have flaws such as not being truly composable and demanding excessive context. It requires substantial upfront input, and each tool invocation consumes considerable context. A comparison with the gh CLI tool reveals that the latter manages context more efficiently. Although feedback suggests MCP could be viable for certain end-user applications like automating specific tasks, it is currently harder to use than traditional coding. The reliance on inference in its current form further undermines its effectiveness in various scenarios.
MCP isn't truly composable; most composition happens through inference, which affects its effectiveness in coding tasks.
MCP consumes more context than simply writing and running code, making it less efficient for tasks.
Even considering future developments, current MCP is still harder to use than traditional coding methods.
For domain-specific applications, even if MCP seems suitable, code generation generally proves to be the better choice.
Collection
[
|
...
]