Retool has added its answer to the question every developer platform is now asked. In Retool MCP vs. CLI, Prajya Sharma lays out two surfaces for building with agents, an MCP server and a CLI, and the metaphor is the part people will remember: “The CLI is you under the sink: you can see the problem directly and fix it right there.” The MCP server is the other room. It is for discovering existing apps and resources, making small targeted changes, admin work like permissions and audit logs, and anything done from a chat-only surface, because “you can’t run the CLI on a chat-only platform like ChatGPT.” The CLI is for building new apps locally with full control, keeping them on disk under Git, and importing external React code.
There are no numbers in the post, so there is nothing to discount, and the reasoning is the kind that holds without them. “As an app build increases in complexity, prompts get more specific, which can make the MCP’s agent-to-agent building less efficient,” which is the honest version of a claim most vendors make the other way round. And the conclusion refuses the premise of the title: “you don’t have to pick one surface and stick with it — most builders will end up using both, just for different parts of the job.” That makes three providers in this series inside two weeks, Postman, GitHub, and now Retool, to answer the same question the same way. The surfaces are projections of one platform for different consumers, and the argument is over.
The catalog maps the two surfaces onto Retool’s record cleanly. The Retool provider page lists 7 API pages, and the MCP server’s discovery and admin work runs over the Retool Apps API, the Retool Resources API, and the Retool Groups API, which is the permissions surface the post hands to MCP. On the Agent Readiness score the MCP server, agent skills, and the full identity layer are lit, delegated identity, protected resource metadata, and dynamic client registration. The agentic access profile maps 17 operations, 10 of them acting.
The Kin Score is 54.7, strong band, carried by contract quality at 69.5 and discoverability at 68.3, with operational transparency at 55.3, access clarity at 48.7, developer ergonomics at 43.5, and contract governance at 31.8. The Agent Readiness score is 44.3, agent-ready. Idempotency, reversibility, and OpenAPI examples are unlit, and the first two matter for exactly the work the post assigns to MCP. Small targeted changes to a production app from a chat window are the calls most likely to be retried, and least likely to be reviewed before they run. Retool has drawn the line between the two surfaces well. The contract under the chat window does not yet say what happens when the same small change is made twice.