Add hybrid search support #43
Conversation
WalkthroughHybrid and vector search capabilities were integrated into the Meilisearch MCP server and client. Documentation was updated to reflect these enhancements. The client and server now accept new parameters for hybrid and vector search. Tests were added to verify the acceptance and schema of these new parameters in the search tool. Changes
Sequence Diagram(s)sequenceDiagram
participant Client
participant MCP_Server
participant Meilisearch
Client->>MCP_Server: search(query, hybrid, vector, retrieveVectors, ...)
MCP_Server->>Meilisearch: search(query, hybrid, vector, retrieveVectors, ...)
Meilisearch-->>MCP_Server: search results
MCP_Server-->>Client: search results
Poem
✨ Finishing Touches
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
CodeRabbit Configuration File (
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (2)
src/meilisearch_mcp/client.py (1)
67-75: Shadowing built-infilteridentifierUsing
filteras a parameter name shadows Python’s built-infilterfunction – not a blocker but easy to trip over in debugging & IDE hints. Consider renaming tofilter_query(and adjust server schema/tests) for clarity.README.md (1)
136-137: Minor punctuation – add comma after parenthetical phrase“with support for hybrid search (combining keyword and semantic search)” →
“with support for hybrid search (combining keyword and semantic search),”🧰 Tools
🪛 LanguageTool
[uncategorized] ~136-~136: Possible missing comma found.
Context: ... tool that can search across one or all indices with support for hybrid search (combini...(AI_HYDRA_LEO_MISSING_COMMA)
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (5)
README.md(4 hunks)pyproject.toml(1 hunks)src/meilisearch_mcp/client.py(2 hunks)src/meilisearch_mcp/server.py(3 hunks)tests/test_mcp_client.py(1 hunks)
🧰 Additional context used
🪛 LanguageTool
README.md
[uncategorized] ~136-~136: Possible missing comma found.
Context: ... tool that can search across one or all indices with support for hybrid search (combini...
(AI_HYDRA_LEO_MISSING_COMMA)
[uncategorized] ~138-~138: Loose punctuation mark.
Context: ...eyword and semantic search): - search: Search through Meilisearch indices with...
(UNLIKELY_OPENING_PUNCTUATION)
🪛 GitHub Actions: Test and Lint
tests/test_mcp_client.py
[error] 689-689: TestHybridSearch.test_search_with_hybrid_parameters failed due to MeilisearchApiError: invalid_embedder. Cannot find embedder with name 'default'. Possibly due to incompatible Meilisearch version.
[error] 724-724: TestHybridSearch.test_search_with_vector_parameter failed due to MeilisearchApiError: bad_request. Unknown field 'retrieveVectors' in search parameters. Possibly due to incompatible Meilisearch version.
[error] 760-760: TestHybridSearch.test_search_semantic_only failed due to MeilisearchApiError: invalid_embedder. Cannot find embedder with name 'default'. Possibly due to incompatible Meilisearch version.
🔇 Additional comments (2)
src/meilisearch_mcp/server.py (1)
227-256: Schema usesretrieveVectors, client fix proposesretrieveVectorAfter correcting the client (see previous comment) this schema must mirror the exact casing and singular form or requests will 400.
Update the property key and description accordingly, then bumprequired/additionalPropertiesrules if necessary.pyproject.toml (1)
6-8: Confirm minimum SDK version truly includes vector API
meilisearch>=0.34.0is necessary but maybe not sufficient if the project relies on very fresh fields (retrieveVector).
Double-check release notes; if first supported in ≥0.35.0, bump here to avoid runtime 400s.
| hybrid: Optional[Dict[str, Any]] = None, | ||
| vector: Optional[List[float]] = None, | ||
| retrieve_vectors: Optional[bool] = None, |
There was a problem hiding this comment.
retrieveVectors parameter name appears incorrect – causes 400-error
CI shows MeilisearchApiError: Unknown field 'retrieveVectors'. In v1.6 (SDK ≥ 0.34) the parameter accepted by the search endpoint is retrieveVector (singular) – see Meilisearch vector-search docs.
Please align with the official name to avoid hard failures.
- retrieve_vectors: Optional[bool] = None,
+ retrieve_vector: Optional[bool] = None,
…
- if retrieve_vectors is not None:
- search_params["retrieveVectors"] = retrieve_vectors
+ if retrieve_vector is not None:
+ search_params["retrieveVector"] = retrieve_vectorRemember to update the server handler, JSON schema and tests accordingly.
Also applies to: 76-82
🤖 Prompt for AI Agents
In src/meilisearch_mcp/client.py around lines 55 to 57 and 76 to 82, the
parameter name retrieveVectors is incorrect and causes a 400 error. Rename the
parameter from retrieveVectors to retrieveVector (singular) to match the
official Meilisearch v1.6 API. Update all references including the server
handler, JSON schema, and tests to use retrieveVector consistently.
There was a problem hiding this comment.
This has been changed since, the correct parameter name is plural
There was a problem hiding this comment.
@tpayet Merci pour la précision !
Je me suis référé à une version antérieure de la doc ; en effet, la 1.6.x / SDK ≥ 0.34 accepte maintenant le paramètre retrieveVectors au pluriel.
Je retire donc mon observation ; le code et le schéma actuels sont corrects. 👍
There was a problem hiding this comment.
lol this robot is nuts. It responds to your English in French 😆
There was a problem hiding this comment.
Seems like the humans are having a chat. I'll hop back into my burrow for now. If you need me again, just tag @coderabbitai in a new comment, and I'll come hopping out!
tests/test_mcp_client.py
Outdated
| {"id": 2, "title": "JavaScript Guide", "description": "Modern JS features"}, | ||
| {"id": 3, "title": "Machine Learning", "description": "AI and ML concepts"}, | ||
| ] | ||
| await simulate_mcp_call( | ||
| mcp_server, | ||
| "add-documents", | ||
| {"indexUid": index_name, "documents": documents}, | ||
| ) | ||
| await wait_for_indexing() | ||
|
|
||
| # Note: This test simulates the API call structure but won't actually | ||
| # perform semantic search without a configured embedder in Meilisearch | ||
|
|
||
| # Test search with hybrid parameters | ||
| search_params = { | ||
| "query": "programming", | ||
| "indexUid": index_name, | ||
| "hybrid": {"semanticRatio": 0.7, "embedder": "default"}, | ||
| "limit": 5, | ||
| } | ||
|
|
||
| # The search should accept these parameters without error | ||
| response = await simulate_mcp_call(mcp_server, "search", search_params) | ||
| response_text = assert_text_content_response(response, "Search results") | ||
|
|
||
| # Even if embedder is not configured, the API should handle the request | ||
| assert "Search results for 'programming'" in response_text | ||
|
|
||
| # Cleanup | ||
| await simulate_mcp_call(mcp_server, "delete-index", {"uid": index_name}) | ||
|
|
||
| async def test_search_with_vector_parameter(self, mcp_server): | ||
| """Test that search accepts vector parameter""" | ||
| # Create test index | ||
| index_name = generate_unique_index_name("vector_search") | ||
| await simulate_mcp_call( | ||
| mcp_server, "create-index", {"uid": index_name, "primaryKey": "id"} | ||
| ) | ||
|
|
||
| # Add test documents | ||
| documents = [{"id": 1, "content": "Test document"}] | ||
| await simulate_mcp_call( | ||
| mcp_server, | ||
| "add-documents", | ||
| {"indexUid": index_name, "documents": documents}, | ||
| ) | ||
| await wait_for_indexing() | ||
|
|
||
| # Test search with vector parameter | ||
| search_params = { | ||
| "query": "test", | ||
| "indexUid": index_name, | ||
| "vector": [0.1, 0.2, 0.3, 0.4, 0.5], | ||
| "retrieveVectors": True, | ||
| } | ||
|
|
||
| # The search should accept these parameters without error | ||
| response = await simulate_mcp_call(mcp_server, "search", search_params) | ||
| response_text = assert_text_content_response(response, "Search results") | ||
| assert "Search results for 'test'" in response_text | ||
|
|
||
| # Cleanup | ||
| await simulate_mcp_call(mcp_server, "delete-index", {"uid": index_name}) | ||
|
|
||
| async def test_search_semantic_only(self, mcp_server): | ||
| """Test semantic-only search with semanticRatio=1.0""" | ||
| # Create test index | ||
| index_name = generate_unique_index_name("semantic_only") | ||
| await simulate_mcp_call( | ||
| mcp_server, "create-index", {"uid": index_name, "primaryKey": "id"} | ||
| ) | ||
|
|
||
| # Add test documents | ||
| documents = [ | ||
| {"id": 1, "content": "Artificial intelligence and machine learning"} | ||
| ] | ||
| await simulate_mcp_call( | ||
| mcp_server, | ||
| "add-documents", | ||
| {"indexUid": index_name, "documents": documents}, | ||
| ) | ||
| await wait_for_indexing() | ||
|
|
||
| # Test semantic-only search | ||
| search_params = { | ||
| "query": "AI ML", | ||
| "indexUid": index_name, | ||
| "hybrid": { | ||
| "semanticRatio": 1.0, # Pure semantic search | ||
| "embedder": "default", | ||
| }, | ||
| } | ||
|
|
||
| response = await simulate_mcp_call(mcp_server, "search", search_params) | ||
| response_text = assert_text_content_response(response, "Search results") | ||
| assert "Search results for 'AI ML'" in response_text | ||
|
|
||
| # Cleanup | ||
| await simulate_mcp_call(mcp_server, "delete-index", {"uid": index_name}) | ||
|
|
There was a problem hiding this comment.
Integration tests fail against vanilla Meilisearch – mock or xfail hybrid/vector paths
The pipeline errors (invalid_embedder, unknown field 'retrieveVectors') show that the test suite expects a server with an embedder configured and the now-incorrect parameter name.
To keep CI green on default installations:
- Rename the param after code fix (
retrieveVector). - Mark the three hybrid/vector tests with
pytest.mark.skipif(or mockMeilisearchClient.search) when the remote server does not advertise the vector-search capability. - Alternatively run an embedded Meilisearch container with a test embedder during CI.
Example:
-@pytest.mark.asyncio
async def test_search_with_hybrid_parameters(self, mcp_server):
+@pytest.mark.asyncio
+@pytest.mark.skipif(
+ os.getenv("MEILI_VECTOR_ENABLED") != "1",
+ reason="Vector search not enabled in CI environment",
+)Committable suggestion skipped: line range outside the PR's diff.
🧰 Tools
🪛 GitHub Actions: Test and Lint
[error] 689-689: TestHybridSearch.test_search_with_hybrid_parameters failed due to MeilisearchApiError: invalid_embedder. Cannot find embedder with name 'default'. Possibly due to incompatible Meilisearch version.
[error] 724-724: TestHybridSearch.test_search_with_vector_parameter failed due to MeilisearchApiError: bad_request. Unknown field 'retrieveVectors' in search parameters. Possibly due to incompatible Meilisearch version.
[error] 760-760: TestHybridSearch.test_search_semantic_only failed due to MeilisearchApiError: invalid_embedder. Cannot find embedder with name 'default'. Possibly due to incompatible Meilisearch version.
🤖 Prompt for AI Agents
In tests/test_mcp_client.py around lines 615 to 765, the integration tests for
hybrid and vector search fail on vanilla Meilisearch due to missing embedder
configuration and incorrect parameter naming. To fix this, rename the parameter
"retrieveVectors" to "retrieveVector" to match the updated API. Then, mark the
three tests involving hybrid and vector search with pytest.mark.skipif to skip
them when the Meilisearch server does not support vector search capabilities, or
alternatively mock the MeilisearchClient.search method in these tests. This will
prevent CI failures on default Meilisearch installations without embedder
support.
There was a problem hiding this comment.
Actionable comments posted: 0
♻️ Duplicate comments (2)
tests/test_mcp_client.py (2)
647-650:⚠️ Potential issueParameter name still out-of-sync with Meilisearch API
The server (≥ v0.34.0) expects
retrieveVector, notretrieveVectors.
Both the schema assertion here and subsequent tests will fail against a compliant server.-assert "retrieveVectors" in properties -assert properties["retrieveVectors"]["type"] == "boolean" +assert "retrieveVector" in properties +assert properties["retrieveVector"]["type"] == "boolean"Update the test payloads accordingly.
717-723:⚠️ Potential issueRename request parameter to match API
Search requests still send the deprecated
retrieveVectorsflag.- "vector": [0.1, 0.2, 0.3, 0.4, 0.5], - "retrieveVectors": True, + "vector": [0.1, 0.2, 0.3, 0.4, 0.5], + "retrieveVector": True,
🧹 Nitpick comments (1)
tests/test_mcp_client.py (1)
630-640: Relax brittlerequiredassertion
hybrid_schema["required"] == ["embedder"]will break if the backend later makes any additional field mandatory.
Safer:-assert hybrid_schema["required"] == ["embedder"] +assert "embedder" in hybrid_schema.get("required", [])
- Update meilisearch dependency from >=0.33.0 to >=0.34.0 for stable AI-powered search - Add hybrid search parameters to search tool schema: - hybrid object with semanticRatio and embedder fields - vector parameter for custom vectors - retrieveVectors parameter to include vectors in results - Update client.py search method to pass new parameters to SDK - Add comprehensive tests for hybrid search functionality - Update README with hybrid search examples and documentation This enables users to leverage Meilisearch's semantic and hybrid search capabilities through the MCP interface, combining traditional keyword search with AI-powered semantic search for more relevant results. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <[email protected]>
Add pytest.mark.skip decorators to tests that require a Meilisearch instance with embedders configured. These tests validate the API structure but cannot run without proper embedder setup.
tpayet
force-pushed
the
feature/hybrid-search-support
branch
from
3f9c39f to
c87bcfc
Compare
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (4)
README.md (4)
136-136: Add comma to clarify hybrid search descriptionIn the sentence, use a comma instead of parentheses to improve readability and consistency:
- The server provides a flexible search tool that can search across one or all indices with support for hybrid search (combining keyword and semantic search): + The server provides a flexible search tool that can search across one or all indices with support for hybrid search, combining keyword and semantic search:🧰 Tools
🪛 LanguageTool
[uncategorized] ~136-~136: Possible missing comma found.
Context: ... tool that can search across one or all indices with support for hybrid search (combini...(AI_HYDRA_LEO_MISSING_COMMA)
152-164: Clarify JSON example commentsThese code blocks include
//comments inside JSON, which is not valid JSON. Consider moving explanatory comments outside the JSON or marking these as pseudocode. This will help users copy/paste examples without errors.
166-175: Clarify JSON example comments in semantic search sectionSimilarly, the semantic search example embeds comments inside the JSON. Either remove the inline comments or wrap examples with a disclaimer (e.g., “Note: JSON comments added for clarity”).
245-246: Unify wording in the Available Tools sectionThe search description uses “single or multiple indices” here, but earlier it says “one or all indices.” For consistency and clarity, align both phrases and match the refined hybrid search description:
- - `search`: Flexible search across single or multiple indices with support for hybrid search (keyword + semantic), custom vectors, and filtering/sorting options + - `search`: Flexible search across one or all indices with support for hybrid search, combining keyword and semantic search, custom vector search, and filtering/sorting options
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (5)
README.md(4 hunks)pyproject.toml(1 hunks)src/meilisearch_mcp/client.py(2 hunks)src/meilisearch_mcp/server.py(3 hunks)tests/test_mcp_client.py(1 hunks)
🚧 Files skipped from review as they are similar to previous changes (4)
- pyproject.toml
- src/meilisearch_mcp/server.py
- src/meilisearch_mcp/client.py
- tests/test_mcp_client.py
🧰 Additional context used
🪛 LanguageTool
README.md
[uncategorized] ~136-~136: Possible missing comma found.
Context: ... tool that can search across one or all indices with support for hybrid search (combini...
(AI_HYDRA_LEO_MISSING_COMMA)
[uncategorized] ~138-~138: Loose punctuation mark.
Context: ...eyword and semantic search): - search: Search through Meilisearch indices with...
(UNLIKELY_OPENING_PUNCTUATION)
🔇 Additional comments (1)
README.md (1)
195-200: New search parameters are well-documentedThe additions of
hybrid,vector, andretrieveVectorsparameters are correctly described with defaults and requirements. Great job updating the parameter list.
Summary
Changes
meilisearch>=0.34.0(from>=0.33.0)hybridobject withsemanticRatioandembedderfieldsvectorparameter for custom vector searchretrieveVectorsparameter to include vectors in search resultsclient.pyto pass new parameters to the Meilisearch SDKTest Plan
Notes
The integration tests require a running Meilisearch instance. Some tests may fail if the instance requires API authentication. The hybrid search features require embedders to be configured in Meilisearch for full functionality.
Related Links
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Documentation
Tests
Chores