What an MCP client can do
Found this helpful? Share it:
Found this helpful? Share it:
idapt exposes one MCP tool, idapt. You call it with a
structured input and address every capability through an action path 🔧
Call the tool with {action: "<resource> <verb>", args: {...}}. The
action names a resource and a verb; the arguments ride in args as plain JSON, with no shell quoting.
The command column below is that action string: a row reading idapt drive read is the call {action: "drive read", args: {...}}. It is the same
grammar the idapt CLI speaks.
The older {cmd: "idapt ..."}string form is not
accepted. Always send a structured action.
Every call except one needs a workspace-qualified handle, written /owner/workspace-slug. There is no chat context to guess
from, so a bare slug is rejected.
The exception is {action: "workspace list"}. Start
there to see the workspaces your key can reach, then use one of those
handles for the rest of your calls.
Ask the tool to describe itself, with no separate call type:
Contract(arguments, types, return shape, errors): {action: "help drive read"}, or args.help: true on the verb.
Playbook(when to use it, why, anti-patterns): {action: "instructions drive"}, or args.instructions: true.
Drive shows the pattern. Every resource works the same way, so browse the full set in the command reference.
| Command | What it does |
|---|---|
idapt drive list | List files (cursor-paginated). Filter to a folder with parent_id. |
idapt drive read | Read a file's text, with paging, tail or a regex search. |
idapt drive get-metadata | Get a file's metadata by id. |
idapt drive upload | Upload a file (multipart). Provide workspace_id or parent_id. |
idapt drive update | Update file name/content. OCC via expected_updated_at. |
idapt drive delete | Trash a file by id (recoverable with `drive restore`). |
idapt drive restore | Restore a trashed file. |
idapt drive permanent-delete | Permanently delete a trashed file (irreversible). |
idapt drive create-folder | Create-or-get a folder (idempotent). |
idapt drive move | Move a file to a new parent_id (null = workspace root). |
idapt drive signed-url | Mint a short-lived direct link to a file's bytes. |
idapt drive content-versions | List a file's content version history (newest first). |
idapt drive restore-version | Restore a file to a previous content version. |
idapt drive glob | Find files by glob pattern (e.g. **/*.ts) across the drive. |
idapt drive grep | Search file contents by regex (modes: content/files/count, ripgrep-style context). |
The server fails closed. There are no always-available actions: every action is scope-gated, and an action with no scope mapping is denied. A few surfaces are excluded from MCP entirely:
Computer use: the desktop screenshot-and-click loop needs live visual feedback.
Background actions and reading buffered tool output: they reference in-flight chat executions.
Sub-agent write verbs: launching and steering workers stays in chat.
Hub actions: installing and publishing are internal, not agent-callable.
Related articles
idapt MCP Server
Connect Claude Code, Cursor, Codex, and any other MCP client to your idapt workspace.
MCP permissions
API key scopes, what each one grants, and how to pick the narrowest set for your workflow.
Connect your tool
Create a scoped API key and point any MCP client at idapt's endpoint, then check that it worked.
Was this helpful?