docstar-cli calls published DocStar API endpoints directly from your terminal, or exposes them as MCP tools for AI clients (Claude Desktop, Claude Code, etc.). Point it at a DocStar docs site, pick which modules you want (e.g. “Slack”), and it turns each published endpoint into a real command you can run.

You can install from multiple docs sites/collections at once — each is kept separate under its own CLI name.

npm install
npm run build
npm link

npm link puts docstar-cli on your PATH. Alternatively, run it without linking via node ./bin/run.js <command> from the repo root.

Everything you install is grouped under three levels:

docstar-cli <cli-name> <module> <command>

• cli-name — identifies the docs site/collection you installed from (e.g. msg91). Set under the collection’s Domain settings in the docs editor; auto-generated from the collection name if not set.

• module — a group of endpoints (e.g. slack).

• command — a single endpoint (e.g. send-message-1).

Installed data lives under ~/.docstar/:

• ~/.docstar/config.json — installed collections/modules, grouped by collection.

• ~/.docstar/modules/<cli-name>/<module>.json — each module’s full endpoint contract.

Discover published modules for a DocStar docs site, pick which ones to install, and register their CLI commands.

Args

• domain (required) — DocStar docs domain, e.g. docs.msg91.com or localhost:3000.

Flags

• --collectionId <id> — Collection id. Required when the domain is not a custom domain (e.g. localhost).

Examples

docstar-cli init docs.msg91.com
docstar-cli init localhost:3000 --collectionId paM4R4A26Hvb

This fetches the published modules for the collection, shows an interactive checklist (all modules pre-checked — press space to toggle, enter to confirm), and for each selected module:

• saves its full endpoint contract to ~/.docstar/modules/<cli-name>/<module>.json

• registers a real CLI command for every endpoint: docstar-cli <cli-name> <module> <command>

Running init again against a different domain/collection adds it alongside any others already installed — nothing gets overwritten.

List installed collections, the modules within one, or every endpoint of one module.

Args (both optional)

• collection — CLI name to show installed modules for, e.g. msg91.

• module — module to show all installed endpoints for, e.g. slack.

Flags

• --all — list every installed collection (default behavior when no arguments are given).

Examples

docstar-cli list                    # every installed collection
docstar-cli list --all              # same as above
docstar-cli list msg91              # modules installed for the "msg91" collection
docstar-cli list msg91 slack        # Slack's installed endpoints in "msg91"

Sample endpoint listing output:

send-message-1
  send message
  POST
  required: message

Calls the actual API endpoint. These commands only exist after you’ve run init and selected the module they belong to — there is no static list; run docstar-cli list to see what’s installed.

Required/optional parameters become flags. Omit a required flag and you’ll be prompted for it interactively.

Examples

docstar-cli msg91 slack send-message-1 --message "hello from the CLI"

# Or omit required parameters and answer the prompt instead:
docstar-cli msg91 slack send-message-1
? Enter value for message

The command prints the response status and body from the actual API call.

Starts an MCP (Model Context Protocol) server exposing every installed endpoint, across all installed collections, as a callable tool for AI clients.

• Each endpoint becomes one tool named <cli-name>__<module>__<command>, e.g. msg91__slack__send-message-1, with required/optional parameters exposed in its schema.

• Run docstar-cli init <domain> first so there’s at least one installed module for mcp to expose.

• This command does not exit on its own — it stays running and communicates over stdio for as long as the connecting AI client keeps the connection open.

Example: register with an MCP client (.mcp.json / claude_desktop_config.json)

{...}{
  "mcpServers": {
    "docstar": {
      "command": "docstar-cli",
      "args": ["mcp"]
    }
  }
}

Once registered, a prompt like “call the slack api and send a message saying hi” resolves directly to the matching tool call — the AI client asks you for any required value it can’t infer, just like any other MCP tool.

These come from the CLI framework (@oclif/plugin-help, @oclif/plugin-plugins) rather than DocStar itself:

• docstar-cli help [command] — show help for the CLI or a specific command.

• docstar-cli plugins — manage oclif plugins.

• docstar-cli --version / docstar-cli -v — print the installed version.

# 1. Install modules from a docs site
docstar-cli init docs.msg91.com

# 2. Check what got installed
docstar-cli list
docstar-cli list msg91
docstar-cli list msg91 slack

# 3. Call an endpoint directly
docstar-cli msg91 slack send-message-1 --message "hello from the CLI"

# 4. Or expose everything installed as MCP tools for an AI client
docstar-cli mcp

• init talks to GET /p/module.json?collectionId=... (collection name, CLI name, and list of modules) and GET /p/<module>/module.json?collectionId=... (a module’s endpoint contract) on the docs site.

• Each endpoint’s contract (method, url, base url, headers, body, and CLI parameter names) is saved locally and used to build the real HTTP request when you run its command — no network round-trip to the docs site is needed once a module is installed.

• Commands are generated as real files on disk under this repo’s dist/commands/<cli-name>/<module>/. Running npm run build wipes dist/, so you’ll need to re-run docstar-cli init for each installed collection afterwards to reinstall your modules.

npm run build   # compile TypeScript to dist/
npm run lint    # eslint
npm test        # mocha