Command
tent mcp
A server of MCP on this computer, for an AI that builds the app with you: it offers the project's commands as tools. Its client starts it, as a program, and speaks to it over its input and its output in the standard's form for such a program, one message on each line. It needs no sign-in. tent mcp approve asks an outside MCP server for its tools and approves them.
Usage
tent mcp [--data <folder>]
tent mcp approve <service> [--as <user id>] [--data <folder>]A client names tent mcp as the program to start in the folder of the project, as { "command": "tent", "args": ["mcp"] }.
tent mcp approve asks an outside MCP server for its tools, through the engine, with the values of the service's auth, or with --as with the connection of that user. It shows each tool with its description and its input, and writes them with their fingerprints into mcp.lock.json in the root of the project, in place of what the file held of that service: { "<service>": { "tools": { "<tool>": { "about", "schema", "fingerprint" } } } }. The build makes each a call of the service. See option.service.mcp.
Options
| Option | Means |
|---|---|
--data <folder> | The data folder of the engine: of tent mcp, the one whose mode it tells when it starts; of tent mcp approve, the one that asks the server. Left out, it is .tent/ of the project |
--as <user id> | Of tent mcp approve: the user whose connection asks a server with oauth |
Output
tent mcp offers these tools:
| Tool | Is | Only reads |
|---|---|---|
tent_map | command.map | yes |
tent_docs | command.docs | yes |
tent_check | command.check | yes |
tent_run | command.run | no |
tent_rows | tent data rows (command.data) | yes |
tent_logs | command.logs | yes |
tent_shot | command.shot | yes |
tent_history | command.history | yes |
The input of each tool is the options of its command, by their names, and its answer is what the command prints with --json: { ok, value, findings }. An input that does not fit the options is the tool's error, code.T0002.
The tools that need an engine reach it as their commands do: through data of their input, or .tent/ of the project. So tent_run acts as the caller its as names, and as the system where it is left out. Beside an engine on live data, tent mcp says so on its error output when it starts.
tent mcp approve with --json prints { "written", "tools" }, each tool by its dotted name.
See also
topic.mcp, topic.commands, command.run, option.service.mcp, code.T7112