Tools Overview

progrok activates two kinds of tool surfaces with the same xAI OAuth session: upstream server-side Grok tools that are passed through the proxy, and local CLI commands that wrap common Grok workflows with better ergonomics.

Proxy tools

When a client calls http://127.0.0.1:18645/v1/*, progrok forwards the request to xAI with your OAuth bearer token. The client can use OpenAI-style request shapes while xAI decides which tools and models your account can use.

Tool	Purpose	How progrok helps
`web_search`	Search current web sources.	OAuth injection lets OpenAI-compatible clients request Grok search tools without xAI OAuth plumbing.
`x_search`	Search public X posts and conversations.	Useful for current market, release, and community research through the same local endpoint.
`code_execution` / `code_interpreter`	Run Python code in an upstream sandbox where available.	Responses API and SDKs may use different aliases; the proxy forwards either shape.
`collections_search` / `file_search`	Search uploaded file collections.	Uses account-level xAI access; management credentials may still be needed for collection administration.
`attachment_search`	Search files attached to a message.	Activated by uploaded file attachments and billed as a separate tool invocation class.
`view_image`	Analyze images discovered during Web Search or X Search.	Useful when search results include screenshots, charts, or product images.
`view_x_video`	Analyze videos discovered during X Search.	Useful for release demos, clips, and public product videos.
Function calling	Let the model call functions you define.	progrok handles auth only; your client still owns function definitions and execution.
Remote MCP	Connect model workflows to external MCP servers where supported.	The proxy keeps the auth boundary local while the client owns the tool topology.

Compatibility caveats

OpenAI Responses MCP fields such as require_approval and connector_id are not part of xAI's current Remote MCP surface.
gRPC paths do not accept every REST alias. In particular, code_interpreter and file_search aliases are REST/SDK conveniences, while canonical xAI names are code_execution and collections_search.
Tool requests bill both model tokens and tool invocations. Web, X, and code tools are priced per 1K calls; attachment and collection search have their own rates.

Local commands

Direct commands exist where raw HTTP is not the best user experience. They use the same stored OAuth credential but add source flags, JSON output, polling, download handling, and machine-readable metadata.

Command	Activated capability	Typical use
`progrok search`	Grok Responses + web/X tools	Research with citations, JSON output, and optional reasoning effort.
`progrok image`	Imagine image generation/editing	Generate from text or edit with one or more reference files.
`progrok video`	Imagine video generation/editing	Submit async jobs, poll progress, and save completed videos.
`progrok models`	Model catalog	Inspect available text, coding, reasoning, image, and video models.
`progrok capabilities --json`	Agent-readable metadata	Let other tools discover ports, commands, models, and direct endpoints.

Tool activation pattern

progrok login
  -> stores local OAuth session

progrok proxy
  -> activates OpenAI-compatible /v1/* surface

progrok search / image / video / models / capabilities
  -> activate task-specific Grok workflows directly

Choosing a path

Use the proxy when another app already speaks OpenAI-compatible HTTP.
Use direct commands when you want a terminal workflow with files, polling, or JSON output.
Use capabilities --json when another agent or script needs to decide what progrok can do.