Qdrant's whole API as 56 MCP tools. Every claim tested against a real server.

Showing 13 of 56

• qdrant_health_checkcoreread-onlyidempotentConfirm the configured Qdrant instance is reachable and responding.
• qdrant_collection_createcoreCreate a collection: either a single unnamed vector (`vector_size` + `distance`), or one or more named vectors (`vectors`, each a full `VectorParams` — size, distance, and optionally its own `multivector_config` for ColBERT-style multi-vectors or `quantization_config`) — exactly one of the two. `sparse_vectors` defines sparse (keyword-style) vectors at creation time. `quantization_config` (scalar/product/binary) and `strict_mode_config` apply to the whole collection.
• qdrant_collection_listcoreread-onlyidempotentList every collection name in the configured Qdrant instance.
• qdrant_collection_infocoreread-onlyidempotentReturn full config and status of one collection.
• qdrant_collection_updatecoreidempotentUpdate optimizer/HNSW/collection/vector params on an existing collection.
• qdrant_collection_deletecoredestructiveidempotentDelete a collection and all its points; a no-op if it doesn't exist.
• qdrant_collection_existscoreread-onlyidempotentCheck whether a collection exists, without raising if it doesn't.
• qdrant_points_upsertcoredestructiveidempotentInsert or replace points (id + vector + payload) in a collection.
• qdrant_points_getcoreread-onlyidempotentRetrieve points by id; unknown ids are simply omitted, not an error.
• qdrant_points_deletecoredestructiveidempotentDelete points by id list or by payload filter — exactly one of the two.
• qdrant_points_scrollcoreread-onlyidempotentPage through all points in a collection, optionally filtered.
• qdrant_points_countcoreread-onlyidempotentCount points in a collection, optionally matching a filter.
• qdrant_querycoreread-onlyidempotentSearch a collection with Qdrant's unified Query API: vector similarity, hybrid fusion, multi-stage reranking, MMR diversity, payload-based score formulas, ordering by a payload field, or random sampling.
• qdrant_query_batchsearchread-onlyidempotentRun multiple independent queries against one collection in a single round trip — each item takes every query mode and tuning parameter of `qdrant_query` (plain vector, fusion over prefetch, MMR, formula, order_by, sample, search_params, score_threshold, offset); `timeout_seconds` and `consistency` apply to the whole batch.
• qdrant_query_groupssearchread-onlyidempotentVector query grouped by a payload field, up to `group_size` hits per group — e.g. the best-matching chunks per source document. Takes every query mode and tuning parameter of `qdrant_query` except `offset` (groups have no paging).
• qdrant_recommendsearchread-onlyidempotentFind points similar to a set of positive examples and dissimilar to a set of negative ones (vectors or point ids) — Qdrant's recommendation API. Requires at least one of `positive`/`negative`; the server rejects an empty request with a clear error. Examples may be dense, multi- or sparse vectors, matching `using`.
• qdrant_recommend_batchsearchread-onlyidempotentRun multiple independent recommend queries against one collection in a single round trip. Each item takes the tuning parameters of `qdrant_recommend`; `timeout_seconds` and `consistency` apply to the whole batch.
• qdrant_recommend_groupssearchread-onlyidempotentRecommend query grouped by a payload field, up to `group_size` hits per group. Takes the tuning parameters of `qdrant_recommend` except `offset` (groups have no paging).
• qdrant_discoversearchread-onlyidempotentRank points by how well they fit a target within positive/negative context pairs (vectors or point ids) — Qdrant's discovery search, a finer-grained alternative to recommend. The target and context may be dense, multi- or sparse vectors, matching `using`.
• qdrant_discover_batchsearchread-onlyidempotentRun multiple independent discover queries against one collection in a single round trip. Each item takes the tuning parameters of `qdrant_discover`; `timeout_seconds` and `consistency` apply to the whole batch.
• qdrant_distance_matrix_pairssearchread-onlyidempotentPairwise distance matrix between a random sample of points: for each of `sample` points, its `limit` closest neighbors among that same sample — returned as a flat list of (a, b, score) pairs.
• qdrant_distance_matrix_offsetssearchread-onlyidempotentSame distance matrix as `qdrant_distance_matrix_pairs`, in a column-oriented shape (offsets into a shared id list + a parallel score array) — more compact for large samples.
• qdrant_payload_setpayloadidempotentMerge fields into the payload of selected points — exactly one of `ids`/`points_filter`. Existing keys not in `payload` are kept. `key` sets a nested sub-field (e.g. `"metadata.author"`) instead of merging at the payload root.
• qdrant_payload_overwritepayloaddestructiveidempotentReplace the entire payload of selected points with `payload` — exactly one of `ids`/`points_filter`. Unlike `qdrant_payload_set`, existing keys not in `payload` are dropped.
• qdrant_payload_deletepayloaddestructiveidempotentDelete specific payload keys from selected points — exactly one of `ids`/`points_filter`. Other keys are untouched.
• qdrant_payload_clearpayloaddestructiveidempotentWipe the entire payload of selected points, keeping their vectors — exactly one of `ids`/`points_filter`.
• qdrant_payload_facetpayloadread-onlyidempotentCount distinct values of a payload field across the collection (or a filtered subset) — e.g. how many points per `city`. `exact` trades speed for an exact count instead of an approximation.
• qdrant_payload_index_createpayloadidempotentCreate a payload index on `field_name`, speeding up filters that use it. `field_schema` can be a simple type name ("keyword", "integer", "float", "geo", "text", "bool", "datetime", "uuid") or a detailed params object (e.g. a `text` index with a specific tokenizer, or a `keyword` index marked `is_tenant`).
• qdrant_payload_index_deletepayloaddestructiveidempotentDelete the payload index on `field_name`.
• qdrant_collection_vector_createpayloadidempotentAdd a new named vector (dense or sparse) to a collection that already has points, without touching them.
• qdrant_collection_vector_deletepayloaddestructiveidempotentRemove a named vector (dense or sparse) from a collection — points keep their other vectors and payload. Same server-version requirement as `qdrant_collection_vector_create`.
• qdrant_points_batch_updatepayloaddestructiveRun multiple point operations (upsert, delete, set/overwrite/delete/clear payload, update/delete vectors) against one collection, in the order given, in a single request. Each item is one of Qdrant's own tagged operation shapes, keyed by operation name.
• qdrant_vectors_updatepayloaddestructiveidempotentReplace the vector(s) of existing points by id — leaves their payload untouched. For a named-vector collection, `vector` is a dict keyed by vector name; for a single unnamed vector, pass a plain vector.
• qdrant_vectors_deletepayloaddestructiveidempotentRemove specific named vectors from selected points, keeping their payload and other vectors — exactly one of `ids`/`points_filter`.
• qdrant_alias_listadminread-onlyidempotentList every alias on the server with the collection it points to.
• qdrant_collection_aliasesadminread-onlyidempotentList the aliases that point to one collection (an empty list if it has none). Fails with a clear error if the collection doesn't exist.
• qdrant_alias_updateadmindestructiveCreate, rename, re-point or delete aliases — the basis of a zero-downtime reindex: build the new collection, then switch the alias that clients use.
• qdrant_collection_cluster_infoadminread-onlyidempotentShard layout of a collection: how many shards it has and, for each local shard, its point count and replica state. Works on a single-node deployment too (one peer, `remote_shards` empty); shard transfers and resharding only ever appear on a distributed deployment. Fails with a clear error if the collection doesn't exist.
• qdrant_collection_optimizationsadminread-onlyidempotentProgress of the background optimizer for a collection — answers "has it finished indexing?" and "why isn't the collection green?".
• qdrant_collection_copyadmindestructiveidempotentCopy points (all vectors — dense, sparse and multivector — and payload) from one collection to another of the same Qdrant instance, entirely server-side: the vectors never pass through the model.
• qdrant_payload_schema_inferadminread-onlyidempotentProfile the payload keys that actually occur in a collection, so filters can be written against real keys instead of guessed ones.
• qdrant_snapshot_createsnapshotsCreate a snapshot of one collection's current state.
• qdrant_snapshot_listsnapshotsread-onlyidempotentList the snapshots stored for one collection.
• qdrant_snapshot_deletesnapshotsdestructiveidempotentDelete a collection snapshot, freeing its disk space on the server — does not touch the live collection.
• qdrant_snapshot_recoversnapshotsdestructiveidempotentOverwrite `collection_name` with the state captured in a snapshot — everything written since that snapshot is lost. Creates the collection if it doesn't exist.
• qdrant_snapshot_downloadsnapshotsread-onlyidempotentConfirm a collection snapshot exists and return where to fetch it from — this tool does not transfer the (potentially huge) snapshot file itself; download it yourself (e.g. `curl`) from the returned `url`.
• qdrant_storage_snapshot_createsnapshotsCreate a snapshot of the whole storage (every collection and server config), not just one collection.
• qdrant_storage_snapshot_listsnapshotsread-onlyidempotentList the full-storage snapshots stored on the server.
• qdrant_storage_snapshot_deletesnapshotsdestructiveidempotentDelete a full-storage snapshot, freeing its disk space.
• qdrant_storage_snapshot_downloadsnapshotsread-onlyidempotentConfirm a full-storage snapshot exists and return where to fetch it from — same caveat as `qdrant_snapshot_download`: this tool does not transfer the file itself.
• qdrant_telemetryobservabilityread-onlyidempotentServer-wide telemetry: build info, per-collection stats, request counters, memory and hardware usage. Not tied to any one collection.
• qdrant_metrics_prometheusobservabilityread-onlyidempotentReturn the URL where Qdrant serves Prometheus-format metrics — this tool does not fetch the metrics themselves (they're plain text, not JSON); point your Prometheus scraper at the returned `url` instead.
• qdrant_quotas_getobservabilityread-onlyidempotentCurrent server-wide resource quotas (memory/disk limits) and actual usage.
• qdrant_quotas_setobservabilityidempotentUpdate server-wide resource quotas. Only the fields you pass are changed; omitted ones keep their current value.
• qdrant_issues_listobservabilityread-onlyidempotentList the issues Qdrant has detected about its own configuration (e.g. a heavily-filtered field with no payload index). **API Beta** in Qdrant itself — the exact shape can change without notice, so this is returned as-is rather than forced into a fixed schema.
• qdrant_issues_clearobservabilitydestructiveidempotentClear all accumulated issues.

core ● on by default

Collection and point CRUD, qdrant_query and a health check. A complete server on its own.

search

Batched and grouped queries, recommend, discover, distance matrices.

payload

Payload editing and indexing, named vectors, ordered batch operations.

snapshots

Backup and restore, per collection and for the whole storage.

admin

Aliases, shard layout, optimizer progress, collection copy, payload profiling.

observability

Telemetry, metrics, quotas, self-diagnosed issues.

1. 01 / 09

   Hybrid search that could not run

   What was claimed

   Hybrid search over dense and sparse vectors works: the query tool’s own description carried an example.

   What was run

   Run that example against a real Qdrant, then try to write a point with a sparse vector through the upsert tool.

   What actually happens

   The example failed ("Conversion between sparse and regular vectors failed"), and the upsert tool rejected every vector that was not a plain list of numbers. The flagship feature could not work, and no integration test covered sparse vectors at all.

   What changed

   One shared vector type for writing and querying, conditional upserts (insert-only, update-only), and integration tests for named, sparse and multivector points.

   Fixed inv2.1.0#32

   Pinned bytest_hybrid_rrf_over_dense_and_sparse_prefetch

2. 02 / 09

   The alias batch that was not atomic

   What was claimed

   Alias operations sent in one call are applied atomically (the first draft of the tool description).

   What was run

   One request with a valid "create alias" followed by a "create alias" on a collection that does not exist, then list the aliases.

   What actually happens

   The request fails, but the first alias is still there. A batch is not rolled back.

   What changed

   The description says so and recommends a single operation to re-point an alias, which is all-or-nothing by definition. An integration test fails if Qdrant ever changes this.

   Fixed inv2.2.0#33

   Pinned bytest_alias_batch_is_not_rolled_back_when_a_later_operation_fails

3. 03 / 09

   Diversity that did nothing

   What was claimed

   MMR re-ranks results for diversity.

   What was run

   Query with MMR, no candidate limit and a limit of 2, and compare it with the plain query.

   What actually happens

   Identical results. The candidate pool defaults to the limit, so MMR has nothing to choose from. And the offset is applied inside those results: offset 1 with limit 2 returns one point.

   What changed

   Both facts are in the tool description and pinned by a test.

   Fixed inv2.4.0#35

   Pinned bytest_mmr_needs_candidates_beyond_limit_and_offset_pages_inside_the_limit

4. 04 / 09

   Tool definitions that ate the context

   What was claimed

   A compact set of tool definitions.

   What was run

   Measure the input schemas that a client loads into the model’s context.

   What actually happens

   571 KB for the whole catalog and 150 KB for the default toolset: every tool carried its own copy of Qdrant’s 16 KB filter schema.

   What changed

   The advertised schemas were compacted by roughly two thirds without changing what is accepted, and a test now enforces a size budget so the weight cannot creep back.

   Fixed inv2.5.0#36

   Pinned bytest_the_default_toolset_stays_within_its_schema_budget

5. 05 / 09found by an independent review

   A compact schema that refused legal calls

   What was claimed

   The compact filter schema accepts exactly what Qdrant’s filter model accepts.

   What was run

   Validate the advertised schema and the real model against the same valid and invalid inputs.

   What actually happens

   The first draft refused a legal single condition where a list was expected. After parity tests were added, an independent review found it also refused explicit nulls and a legal "global" option. A client validating arguments against the schema would have rejected legal calls.

   What changed

   Both fixed. The parity tests now run the schema and the model against the same samples, and a second test fails if the installed client gains a condition the hand-written grammar does not mention.

   Fixed inv2.5.0v2.6.0#36#37

   Pinned bytest_filter_schema_accepts_what_qdrants_filter_accepts

6. 06 / 09found by an independent review

   The filter that deleted everything

   What was claimed

   A delete needs exactly one selector, ids or a filter, so it cannot be ambiguous.

   What was run

   Delete with the filters {}, {"must": []} and {"should": []} on a small collection.

   What actually happens

   Each one deleted every point. An empty filter matches everything, and nothing in the tool said no. A model building its condition list dynamically and ending up with an empty one would wipe the data.

   What changed

   A filter with no conditions is refused in delete, payload and vector tools and inside batch updates, and a batch with one refused operation sends nothing at all.

   Fixed inv2.6.0#37

   Pinned bytest_an_empty_filter_no_longer_wipes_the_collection

7. 07 / 09found by an independent review

   Atomic, since the third phase, and never true

   What was claimed

   The batch-update tool runs its operations atomically (in its description since the third phase).

   What was run

   A valid upsert followed by one with the wrong vector size, then count the points.

   What actually happens

   The first one stays applied. It was never atomic. This is the same mistake as the alias entry: a property of Qdrant asserted without running it.

   What changed

   The description, the tool title, the README and this page no longer say it, and a test pins the real behavior.

   Fixed inv2.6.0#37

   Pinned bytest_a_batch_is_not_all_or_nothing_as_the_tool_now_says

8. 08 / 09found by an independent review

   A filter that silently did not filter

   What was claimed

   A prefetch stage takes a filter.

   What was run

   Give a stage Qdrant’s own key, "filter", instead of the tool’s "query_filter".

   What actually happens

   No error, and the results were not filtered: unknown keys were silently ignored. A misspelled option of MMR ran with its defaults the same way, and a typo in a point’s payload key dropped the payload.

   What changed

   Nested inputs now reject unknown keys with a clear error.

   Fixed inv2.6.0#37

   Pinned bytest_an_unknown_key_in_a_nested_input_is_an_error_not_a_silent_no_op

9. 09 / 09found by an independent review

   Keys that could not be pasted into a filter

   What was claimed

   Payload profiling reports keys written the way Qdrant’s filters expect.

   What was run

   Profile a collection with a key that contains a space, then use the printed key in a filter.

   What actually happens

   Qdrant answers 400: such a key needs double quotes. A literal key "k[]" was read as an array path, and an index created on a quoted key never matched its entry in the profile.

   What changed

   Keys are quoted where the syntax requires it, which is also the form Qdrant reports for indexes, so keys now match their index.

   Fixed inv2.6.0#37

   Pinned bytest_schema_infer_emits_keys_that_work_as_filter_keys_and_index_keys