DSers MCP¶
Dropshipping is two businesses wearing one storefront: selling to a customer, and buying from a supplier. The selling side is where the chat window has been useful so far. This is the buying side. DSers is the ordering platform between AliExpress suppliers and a seller's store, and its official MCP endpoint exposes supplier product discovery, listing preparation, supplier order placement and shipment tracking to an assistant.
Server type: Remote (Streamable HTTP)
Endpoint: https://ai.dsers.com/mcp/mcp-so
Auth: OAuth 2.1 with PKCE, no API keys
Scopes: 23, from product:import to order:write and tracking:read
Registry: the mcp.so listing is marked verified by the vendor account
Protocol: 2025-11-25 advertised on the refusal
What the endpoint says about itself¶
A keyless initialize answers 401 with a machine-readable refusal rather than a login page:
{"error":"invalid_request","error_description":"Authorization required. This endpoint uses OAuth 2.1 + PKCE (no API keys).",
"mcp_protocol_version":"2025-11-25",
"oauth_metadata":"https://ai.dsers.com/.well-known/oauth-authorization-server",
"protected_resource_metadata":"https://ai.dsers.com/.well-known/oauth-protected-resource/mcp/mcp-so"}
That is a well-built gate, and it is also the evidence tier this entry stands on: live and gated by the vendor, documented from its own metadata, not driven. Two details are worth recording because they are better practice than most of this catalogue sees. The refusal names a protocol version (2025-11-25) newer than the 2025-06-18 most servers still advertise, and it carries a correlation_id so a support ticket can be matched to a failed handshake.
The authorization server document is public and complete: authorization code with PKCE over S256, refresh tokens, dynamic client registration at /oauth/register, RS256 identity tokens with a published JWKS, and the issuer on the same host.
The scope list is the product document¶
Nobody at a dropshipping company has to guess what this connector can do, because the vendor publishes 23 scopes, and they describe the workflow more precisely than a feature page would:
user:read/write store:read/write theme:read/write
staff:read/write supplier:read
product:read/write product:import product:push
product:mapping order:read/write tracking:read
settings:read/write notification:read billing:read plan:read
Read as a sequence, that is the dropshipping loop: find a supplier product (supplier:read), import it into the store's catalogue (product:import), map the supplier variant to the store variant (product:mapping), publish it (product:push), and then, once it sells, place the supplier order (order:write) and read the tracking that comes back (tracking:read). A seller can grant product:read alone and let an assistant analyse the catalogue without letting it touch an order, which is the right default and is only possible because the scopes are this fine-grained.
The class it enters, and the distinction that earns the entry¶
This catalogue already carries a dropshipping entry: Hustle Got Real lists 170 tools over a dropshipping back office, covering listings, stock, pricing rules, orders and auto-ordering behind one broad hgr scope. It is the store's own nervous system. DSers is the other half of the same business: the supplier relationship, the AliExpress product catalogue, the purchase order that has to be placed with a supplier and the tracking that comes back from it. A seller running a store on one side and a supplier on the other needs both, and the reason to record this one is that the purchase path is where an assistant can actually save a day of clicking.
The rest of the neighbourhood is different in kind: Sellfy is a storefront, WordPress.com is a site builder, and neither places a supplier order.
What it cannot tell you¶
There is no tool list behind the 401, so this guide cannot quote tool names or schemas, and a reader should treat the scope names as the contract rather than as the interface. Pricing is not published on the endpoint and is not repeated here. The store platform matters: the tags name Shopify, and a seller on another platform should confirm the order path before connecting. And the fundamental caution of this category applies in full: an agent with product:push and order:write can publish a wrong product and spend real money on a real supplier, so the scopes should be granted by task, and the catalogue price should be read by a person before the first push.
Frequently asked questions¶
Can an assistant place a supplier order on its own?¶
order:write is in the scope list, so the surface exists, and whether it executes depends on the scope granted at consent. The safe pattern is to grant product:* first, let the agent do catalogue work, and add order:write only after the mapping has been checked on a sample.
Why does the endpoint advertise a newer protocol version?¶
It answers 2025-11-25 on the refusal, which is ahead of the 2025-06-18 that most servers here still negotiate. A client that only speaks the older revision should still connect, since MCP negotiates down, but the header is worth knowing when debugging a handshake.
Is there an API key path?¶
No, and the vendor says so in the refusal: OAuth 2.1 with PKCE only. That removes the usual leak vector of a long-lived key pasted into a config file.
What does the connector need from the seller?¶
A DSers account. The consent screen is the vendor's own, and the scope list above is what a seller is asked to approve, which is the part to read before pressing allow.
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.