Connect an AI assistant (MCP)

StoreRocket runs a remote Model Context Protocol server. An AI assistant connected to it can put your store locator on a website, add and update locations, and read your locator’s analytics, using the same API token and permissions as the REST API.


Server URL	`https://storerocket.io/mcp`
Transport	Streamable HTTP
Authentication	Sign in (OAuth) or `Authorization: Bearer YOUR_TOKEN`

Create a token

Open API Tokens and create a token for your assistant. Tick what it may do:

Permission	Lets the assistant
See locations (`location:read`)	Find and read locations and categories, get the install code
Add locations (`location:create`)	Add locations
Edit locations (`location:update`)	Change locations
Delete locations (`location:delete`)	Delete locations, one per call
See analytics (`analytics:read`)	Read analytics: views, searches, searches that found nothing, clicks, leads

The token reaches the projects your user account belongs to, on a plan that includes the REST API. Revoke it from the same page at any time.

Connect your client

claude mcp add --transport http storerocket https://storerocket.io/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

{
  "mcpServers": {
    "storerocket": {
      "url": "https://storerocket.io/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Any MCP client that can send an Authorization header to a Streamable HTTP server can connect the same way. Keep the token out of shared repositories: in Cursor, use a user-level ~/.cursor/mcp.json. Clients that support MCP sign-in (OAuth 2.1), such as Claude, ChatGPT and Claude Code, need only the server URL. Add https://storerocket.io/mcp without a header; the client opens a StoreRocket page where you log in (or sign up) and press Allow. Allowing lets the assistant see, add and update your locations, read your analytics and get your install code, never your billing, plan, password or team.

Claude Code

claude mcp add --transport http storerocket https://storerocket.io/mcp

Then run /mcp in Claude Code and choose Authenticate.


Discovery	`https://storerocket.io/.well-known/oauth-protected-resource/mcp`
Client registration	Dynamic (`/oauth/register`), public clients; callback addresses of supported assistants or `http://localhost/callback` on any port
Authorization	Authorization code with PKCE `S256` only (`plain` is refused); scope `mcp:use`
Access token	1 hour, refreshed for 90 days of use

To disconnect an assistant, open API Tokens: Connected assistants lists each one with Disconnect. Resetting your password disconnects every assistant.

Tools

Start with list-projects: every other tool takes a project_id from it (a 10-character id such as Xn0pj9v86V). Location tools take a location_id from list-locations.

Tool	Title	What it does	Permission
`list-projects`	Your store locators	Your projects: id, name, location count, status, plan	none
`get-install-code`	Install code	The code that puts the locator on a website, placement steps, and whether it is live (last page seen in 30 days)	See locations
`list-locations`	Find locations	Up to 100 per page, filtered by name, address, city, state, country or postal code; `include` adds hours, filters, fields or calls_to_action	See locations
`get-location`	Location details	One location with hours, categories, custom fields and buttons	See locations
`list-filters`	Location categories	Category (filter) names with how many locations carry each	See locations
`create-locations`	Add locations	Add 1 to 50 locations; the answer lists created and refused, with reasons	Add locations
`update-locations`	Update locations	Change 1 to 50 locations, each `{location_id, changes}`; only the fields sent change, as PATCH does	Edit locations
`delete-location`	Delete a location	Delete one location, permanently	Delete locations
`get-analytics-overview`	Locator performance	Views, searches, searches with no results, clicks and leads for a date range, next to the period before	See analytics
`get-top-searches`	Top searches	The most searched places and words, with how many found nothing	See analytics
`get-searches-with-no-results`	Unmet demand	Where visitors searched and found no location nearby	See analytics
`get-most-clicked-locations`	Most clicked locations	The most clicked locations, by kind of click (opened, directions, calls, emails, website, buttons)	See analytics

Every tool declares its title, an output schema, and whether it reads, writes or deletes (read-only, destructive, idempotent and open-world hints). Location tools call the REST API: the same validation, geocoding and error messages. Analytics tools return the same numbers as Analytics in the dashboard. Dates are YYYY-MM-DD in UTC, up to 366 days per request, 30 days by default. Analytics count what visitors did on the locator, not sales or store visits.

Prompts

Prompt	What it asks the assistant
`weekly-demand-review`	Review the last 7 days: the numbers against the week before, unmet demand, the most clicked locations, and a few actions
`install-store-locator`	Put the locator on a page with `get-install-code`, then call it again to confirm it is live

Limits and errors

Requests share your REST API rate limit. Analytics tools also allow 30 requests a minute per person.
A refused call returns an MCP tool error with the reason, such as a missing permission, a plan without API access, or a project id that is not in your account. Another account’s projects and locations are never returned.
create-locations and update-locations save each item on its own. When none could be saved, the error still lists every item and its reason.
create-locations without coordinates looks each address up while the request waits. For hundreds of rows, use an import or Google Sheets sync in the dashboard.

Start here

AI assistants

Integration guides

Account

Locations

Existing integrations

Connect an AI assistant (MCP)

Create a token

Connect your client

Tools

Prompts

Limits and errors

​Create a token

​Connect your client

​Sign in instead of a token

​Tools

​Prompts

​Limits and errors

Create a token

Connect your client

Sign in instead of a token

Tools

Prompts

Limits and errors