Read-only. Every tool retrieves and analyzes data. Nothing you do through the MCP server creates, modifies, or deletes data in Perform.AI.
Availability
In ChatGPT, adding the Perform.AI plugin takes one click. Until the Claude directory listing is approved, manual setup takes about a minute.
What you can do
Explore and understand your Shipments. Search and filter your organization’s Shipments by carrier, event status, destination, dates, tags, and dozens of other fields, or replay a view you’ve already saved in the Portal.“Show me our shipments imported in the last 30 days that are still in transit with DHL Express, broken down by destination country.”Deep-dive into a specific Shipment. Open one Shipment by its ID and walk through its event history, milestones, and analytics fields.
“Pull up shipment ORDER-88231 and walk me through its checkpoint events. Where did it get delayed?”Analyze delivery performance. Compare carriers and destinations by Shipment counts, event status, and delivery dates.
“Which carriers had the most delayed deliveries to Germany in July 2026?”Analyze your carrier setup. Review the carriers configured for your organization, compare them against Perform.AI’s master carrier catalog, and get recommendations you can then apply in the Portal.
“List our active carrier configs and check which of those carriers support proof-of-delivery tracking.”The full tool catalogue is on the tools page.
How it works
- Server URL:
https://mcp.perform.ai/mcp(Streamable HTTP). Use this URL for manual setup in any client. - ChatGPT plugin URL:
https://mcp.perform.ai/mcp-invoke. This is the URL the ChatGPT plugin listing shows. It’s the same server, with the same data, permissions, and sign-in, but it exposes a compact set of eight tools that the assistant uses to reach the same read operations. See Tools in the ChatGPT plugin. - Authentication: OAuth 2.1 with dynamic client registration. Your client opens a Perform.AI sign-in page. Sign in with your Perform.AI username and password, or with Sign in with Google. No API keys to create or paste.
- Your permissions, nothing more: access through the MCP mirrors your access in the Portal. You only see data for organizations you belong to.
Organization scoping
If you belong to more than one organization, you choose which one the session acts on. The assistant asks you when it needs an organization, and you can switch at any time by asking it to list your organizations and select another. Every organization-scoped tool operates on the currently selected organization, resolved server-side from your session. No tool accepts an organization parameter, and data belonging to organizations you haven’t selected, or don’t belong to, is unreachable. This is enforced on the server, not by the AI client.Limits and behavior
- Read-only: no tool creates, updates, or deletes data.
- Pagination: Shipment searches return at most 10 rows per call; carrier and config listings default to 25 per page. Ask the assistant to continue to the next page rather than requesting very large pages. For bulk exports, use the Portal’s Shipment Overview report instead.
- Date windows: Shipment queries apply a date range to a chosen lifecycle date (import date by default). When comparing filter values with search results, keep the same window and date field.
- Validation over silence: invalid filter fields and values are rejected with an error naming the problem. An empty result means no matching data, not a swallowed mistake.
- No cross-organization access: identifiers belonging to other organizations return not-found; responses never confirm the existence of another organization’s data.
Data handling and privacy
- Authentication uses your Perform.AI credentials; the server never exposes tokens or credential values through tool responses.
- Credential and integration secrets in carrier configurations are never returned, only metadata about them.
- All data access is scoped to your selected organization and your existing Perform.AI permissions.
- Privacy policy: perform.ai/privacy
FAQ
Can the assistant change my carrier configuration or shipments? No. Every tool is read-only. It can analyze your setup and suggest changes, but you apply changes yourself in the Perform.AI Portal. Why does the ChatGPT plugin show a different URL and fewer tools? The plugin connects tohttps://mcp.perform.ai/mcp-invoke, which exposes eight tools instead of one tool per operation. The assistant finds the operation it needs with list_methods and runs it with invoke_method_read, so you get the same data. See Tools in the ChatGPT plugin.
Why does ChatGPT list a write tool for Perform.AI?
ChatGPT lists Select organization under Write tools because it changes which organization your session acts on. It doesn’t modify any data in Perform.AI.
Can I use this with an organization I don’t belong to?
No. Organization membership is validated server-side at selection time, and every subsequent call is scoped to that validated selection.
Which sign-in methods are supported?
Perform.AI username/password and Sign in with Google. Additional SSO options available on the Portal login are planned for the MCP sign-in page.
Which AI clients are supported?
ChatGPT, through the Perform.AI plugin, and any MCP-compatible client over Streamable HTTP. The Claude directory listing is under review; until it’s approved, use manual setup. See availability.
Does the assistant get the full shipment record, including POD and label downloads?
Detail lookups can return the full record (addresses, line items, costs, milestones, analytics). Document download links for PODs and labels are metadata-only through the MCP: retrieve the files through the Documents API or the Portal.
Can I pull large datasets or build reports?
The MCP is built for conversational analysis, not bulk export. Shipment searches page at 10 rows per call, while count and metrics tools answer aggregate questions (top carriers, bookings by status, invoice cost by carrier) without downloading rows. For bulk data, use the Portal’s Shipment Overview report.
Are the Portal’s pre-calculated analytics available?
Per-shipment analytics fields (transit time, estimated delivery date) come with shipment details, and pre-aggregated booking metrics and invoice statistics are available as tools. Dashboard chart KPIs beyond these aren’t exposed today.
Can I connect my own AI agent?
Yes. Any MCP-compatible client over Streamable HTTP can connect, including agents you build. See Installation.
Is the MCP server included in my plan?
Yes. The MCP server is included for organizations with a valid Perform.AI subscription.
Do I need API credentials or a developer to set it up?
No. You sign in with your Perform.AI username and password (or Google) through the client’s built-in OAuth flow. There are no API keys to create or paste, and setup takes about a minute per client.
Is my data used to train AI models?
The MCP server returns data only to the client you connected, over your authenticated session. How the assistant handles that data, including any training use, is governed by your agreement with that AI provider, not by Perform.AI.
Can the assistant see the credentials stored in my carrier configurations?
No. Credential and integration secrets are never returned through any tool, only metadata about them.
Why does the assistant see fewer shipments than the Portal?
Shipment queries apply a date window to one lifecycle date (import date by default). Align the window and date field with your Portal view, or ask the assistant to replay the Portal view itself from a bookmark or a pasted Portal URL.
Can my whole team use it?
Yes. Each person signs in with their own Perform.AI account and sees exactly what their Portal permissions allow. In ChatGPT workspaces, an admin can make the plugin available to everyone in the workspace.
Changelog
Installation
Connect ChatGPT, Claude, Copilot, or your own MCP client.
Tools
Every tool, filter behavior, and workflow guide, including the ChatGPT plugin tools.