Meilisearch September 2026 updates

Welcome to the September 2026 edition of Meilisearch updates! This one covers a busy summer: a native MCP endpoint, boost and hide controls for search rules, faster search, and easier AI setup in Meilisearch Cloud.

Watch: AI search, discovery, and recommendations from one index (IBC2026)

Quentin de Quelen, our CEO and co-founder, took the Future Tech Stage at IBC2026, the International Broadcasting Convention, in Amsterdam on September 11. He demoed live how AI-powered search, discovery, and recommendations can all run from a single index, on real media catalogs. In this session:

• How hybrid search combines full-text and semantic matching, so users always get a relevant result
• How search results and personalized recommendations are served from the same index
• Retrieval in under 50 ms on typical workloads

Watch the talk on YouTube

Product updates: what's new at Meilisearch

Meilisearch now speaks MCP (experimental)

If you're building AI agents or working with AI coding assistants, your search data is now one connection away. Instead of writing and maintaining your own integration layer, you can connect an MCP-compatible tool to Meilisearch and let it query your indexes directly, using an API key you control.

Meilisearch v1.54 introduces a native Model Context Protocol (MCP) endpoint at /mcp.

• In the Cloud: turn on the MCP route from the experimental features section of your project settings.
• Self-hosted: enable the mcpRoute experimental feature.
• Compatibility: the endpoint uses the latest MCP protocol version (2026-07-28), so make sure your client is up to date.
• Authentication: the route uses bearer authentication with your API key. Clients that require OAuth aren't supported.

Read the MCP guide

Boost, demote, and hide results with search rules (experimental)

For e-commerce and content teams, relevance isn't only about matching the query. It's also about business priorities. Boost, demote, and hide give you a direct way to shape results: push seasonal items up, keep out-of-stock products from crowding the first page, or pull discontinued items entirely, without changing your ranking rules for every query.

Meilisearch v1.54 adds a new scale action, available in the Cloud UI and via the API. It adjusts the relevancy of documents selected by a filter or by IDs:

• A weight above 1.0 boosts them, for example to promote a seasonal collection.
• A weight below 1.0 demotes them, for example to push out-of-stock products down.
• A weight of 0.0 hides them from results, for example to remove discontinued items.

Here's how to demote out-of-stock products:

This builds on top of a summer of search rules upgrades:

• Filter-based activation (v1.51): rules can trigger based on the filters in a search request.
• Built to scale (v1.50): up to 75,000 rules with no impact on search.

More to come soon.

Getting started with search rules · v1.54.0 release notes

Search just got faster (again): v1.49 and v1.51

If your documents have many distinct fields, like large product catalogs with varied attributes, or you rely on long synonym lists, these two releases are for you. Both improvements apply once you upgrade, with nothing to configure.

• Fewer disk reads (v1.51): Meilisearch now reads data from disk a single time across the whole search pipeline. The biggest gains are on datasets with many distinct fields. We measured search up to 5.4x faster on a dataset with more than 14,000 fields.
• Faster synonyms (v1.49): Synonyms are now loaded only when a query word matches one. Indexes with large synonym lists can see search up to 13x faster, and you no longer need to move synonyms into documents as a workaround.

Set up embedders in a few clicks

Setting up semantic or hybrid search used to start with writing embedder settings by hand. The new embedders UI turns that into a guided setup, so teams exploring AI-powered search can get to a working configuration with their provider of choice faster, without hand-writing JSON.

Click into any index and open the new AI Embedders page to get started.

Upgrade your disk performance, self-serve

Large or indexing-heavy projects can be limited by disk throughput. The new usage charts show how close you are to those limits, so you can base the upgrade decision on your own data, and act on it without contacting support.

Add the disk performance add-on yourself from the Infrastructure screen or your project settings. The usage charts track your current IOPS and throughput.

Disk performance pricing

More Cloud quality-of-life improvements

Fewer chores for teams running on Meilisearch Cloud. Automatic upgrades mean smaller projects stay on recent versions without anyone having to schedule an upgrade, and team deletion lets admins clean up after a reorganization or a finished project.

• Automatic version upgrades: Cloud projects can now upgrade to new Meilisearch versions automatically. It's on by default for XS, S, and Build plans. Larger instances can switch it on from the project settings page.
• Delete teams: Organization and team owners can now delete teams. Every organization keeps at least one team.

Foreign filters now work across shards: v1.50 and v1.53

If you've split a large dataset across several instances, you no longer have to choose between scale and filtering on related data. Foreign filters, which let you filter documents using data from a related index, now work across your whole network, and the higher limit makes them practical for larger related sets.

• v1.53: Foreign filters now work in sharded setups, retrieving documents across your network. The number of documents a foreign filter can retrieve rose from 100 to 1,000.
• v1.50: The document fetch routes now retrieve documents from every shard. To keep the old local-only behavior, set useNetwork: false.

Dumpless upgrade is now stable: v1.51

For self-hosted teams, upgrading no longer has to mean exporting a dump and reimporting your data. Now that dumpless upgrade is stable, you can rely on it in production and move to new versions with less operational work.

The flag --experimental-dumpless-upgrade is now --upgrade-db, with exactly the same behavior. We also removed three outdated experimental flags: --experimental-replication-parameters, --experimental-no-edition-2024-for-dumps, and --experimental-no-snapshot-compaction.

v1.51.0 release notes

Clearer performance details in multi-search: v1.54

When a multi-search request is slow, the first question is which query is responsible. For developers debugging federated search, performanceDetails now breaks down timing for each query, so you can find the slow one directly instead of guessing.

Debug search performance

Community and ecosystem

New demos in the docs

Looking for inspiration? We've just expanded our demo library. New examples include HackerSearch, Talk with Lex, Paperscope, Judilibre (legal search), and Agentic movie search.

Explore all demos

Want the full details? Read all the Meilisearch release notes on GitHub.