Which PIM and Catalog Tools Offer an MCP Server? (2026)
A dated comparison of PIM and catalog MCP servers, including whether assistants can read or change product data and how each connection is delivered.
Several PIM and catalog platforms now document an MCP server, but the label hides material differences.
Some connections let an assistant search a catalog. Others can create, update or delete records. Some are hosted by the vendor; others run locally. The useful buying question is not simply "Does it support MCP?" It is "What can our assistant do through it, with whose permissions, against what product structure?"
This comparison was checked against vendors' own documentation and announcements on 27 October 2026.
Current published position
| Platform | Published MCP position | Documented scope |
|---|---|---|
| Akeneo | Hosted MCP server | Read, create or update products, product models, families, attributes, categories, assets and workflows. Availability may vary by package.[1][2] |
| Bluestone PIM | Hosted MCP server | Read and manage products, variants, attributes, categories, assets, relations and completeness as the signed-in user.[3] |
| Sales Layer | Hosted MCP server | Catalog and DAM access with read-only and full-access profiles; the latter can create, update and delete.[4] |
| Pimcore | MCP through Datahub and the Pimcore Labs Agent Bundle | Simple REST Datahub documents search, browse and fetch access. Separately, the Agent Bundle documents read and write-capable MCP tools, human-review controls and configuration-dependent direct writes. Confirm the installed components and availability.[5][6] |
| inriver | MCP server that runs locally | Entity queries and developer assistance for API scripts; the published description does not document direct write-back through the MCP tool.[7] |
| Shopify | Storefront and catalog MCP services | Product discovery and lookup for shopping agents, rather than PIM administration.[8] |
| Furniture Connect | Hosted remote MCP server | Permission-dependent tools across products, variants, materials, imagery, workflows and quotes.[9] |
| Salsify | MCP layer announced | Announced for approved product data; no detailed public setup documentation found in this review.[10] |
| Centric, including Contentserv | MCP announced | Named in a 2026 release; public scope and availability were not clear in the reviewed material.[11] |
| Plytix | No public MCP documentation found | This is an absence of documentation as of the review date, not proof that no private or planned capability exists. |
| Syndigo | No public MCP documentation found | Same qualification: check with the vendor for its current position. |
This market will move. Use the source links as the starting point for a current vendor check, not as a permanent certification.
Read access is not a lesser version of MCP
For sales and customer service, read access may be the right design.
A rep can ask which sofas fit a 220-centimeter wall or whether the corner unit is available in a performance fabric. The assistant can answer without being able to rewrite the product record.
Write access suits different work: applying approved corrections, filling supported fields from verified supplier material or creating a quote. It also expands the effect of a mistake. One bad mapping can be repeated across a filtered range.
Before enabling writes, check that permissions follow the user, the changed records are visible and the team can review or reverse the work.
Furniture is a useful stress test
A demo that searches for a product by name proves very little.
Ask the assistant for a precise sellable version: a two-and-a-half-seater, in a named fabric, with a particular leg finish, below a stated width. Then ask it for the correct image and current price.
That test exposes whether the catalog holds variants, materials, dimensions and assets as connected records. A server sitting above a flat product table can return data quickly without resolving the version the customer intends to buy.
Five questions to ask each vendor
- Is the MCP server hosted by you or operated by us?
- Which tools read, create, update or delete data?
- Do permissions follow the signed-in person or a shared API credential?
- Which package, plan or administrator setting is required?
- How do we inspect and reverse changes made through the assistant?
Add a sixth for furniture: can the assistant identify the exact variant, its materials, dimensions and imagery without interpreting a free-text description?
Where Furniture Connect differs
Furniture Connect's MCP server works against a catalog designed for furniture products, variants, materials and finishes. Its live tools can also work with quotes and supported creative workflows, within the user's permissions and entitlements.[9]
That puts the connection close to commercial work. An assistant can move from a customer brief to the catalog records needed for a draft quote, rather than ending at a product-search result.
People still own the decisions. Missing specifications remain missing. Generated imagery needs review, and saving it to a named variant requires authorization. The point is to make more of the maintained catalog available to the people and assistants serving customers, not to turn governance off.
Sources
- Akeneo: Akeneo MCP server overview
- Akeneo: Akeneo MCP server getting started
- Bluestone PIM: Bluestone PIM MCP overview
- Sales Layer: Sales Layer MCP server
- Pimcore: Simple REST Datahub MCP server
- Pimcore: Pimcore Agent Bundle
- inriver: inriver MCP
- Shopify: Storefront MCP
- Furniture Connect: Furniture Connect vs Claude
- Salsify: SalsifyIQ announcement
- Centric Software: Centric AI announcement
