chore: adds field embeddings validation for quantization "none" and warn when vectorSearch is not configured correctly

src/common/config.ts

@@ -319,7 +319,7 @@ export function warnAboutDeprecatedOrUnknownCliArgs(
     if (knownArgs.connectionString) {
         usedDeprecatedArgument = true;
         warn(
-            "The --connectionString argument is deprecated. Prefer using the MDB_MCP_CONNECTION_STRING environment variable or the first positional argument for the connection string."
+            "Warning: The --connectionString argument is deprecated. Prefer using the MDB_MCP_CONNECTION_STRING environment variable or the first positional argument for the connection string."
         );
     }
 
@@ -333,15 +333,15 @@ export function warnAboutDeprecatedOrUnknownCliArgs(
         if (!valid) {
             usedInvalidArgument = true;
             if (suggestion) {
-                warn(`Invalid command line argument '${providedKey}'. Did you mean '${suggestion}'?`);
+                warn(`Warning: Invalid command line argument '${providedKey}'. Did you mean '${suggestion}'?`);
             } else {
-                warn(`Invalid command line argument '${providedKey}'.`);
+                warn(`Warning: Invalid command line argument '${providedKey}'.`);
             }
         }
     }
 
     if (usedInvalidArgument || usedDeprecatedArgument) {
-        warn("Refer to https://www.mongodb.com/docs/mcp-server/get-started/ for setting up the MCP Server.");
+        warn("- Refer to https://www.mongodb.com/docs/mcp-server/get-started/ for setting up the MCP Server.");
     }
 
     if (usedInvalidArgument) {
@@ -372,6 +372,24 @@ export function registerKnownSecretsInRootKeychain(userConfig: Partial<UserConfi
     maybeRegister(userConfig.username, "user");
 }
 
+export function warnIfVectorSearchNotEnabledCorrectly(config: UserConfig, warn: (message: string) => void): void {
+    const vectorSearchEnabled = config.previewFeatures.includes("vectorSearch");
+    const embeddingsProviderConfigured = !!config.voyageApiKey;
+    if (vectorSearchEnabled && !embeddingsProviderConfigured) {
+        warn(`\
+Warning: Vector search is enabled but no embeddings provider is configured.
+- Set an embeddings provider configuration option to enable auto-embeddings during document insertion and text-based queries with $vectorSearch.\
+`);
+    }
+
+    if (!vectorSearchEnabled && embeddingsProviderConfigured) {
+        warn(`\
+Warning: An embeddings provider is configured but the 'vectorSearch' preview feature is not enabled.
+- Enable vector search by adding 'vectorSearch' to the 'previewFeatures' configuration option, or remove the embeddings provider configuration if not needed.\
+`);
+    }
+}
+
 export function setupUserConfig({ cli, env }: { cli: string[]; env: Record<string, unknown> }): UserConfig {
     const rawConfig = {
         ...parseEnvConfig(env),
@@ -392,6 +410,7 @@ export function setupUserConfig({ cli, env }: { cli: string[]; env: Record<strin
     // We don't have as schema defined for all args-parser arguments so we need to merge the raw config with the parsed config.
     const userConfig = { ...rawConfig, ...parseResult.data } as UserConfig;
 
+    warnIfVectorSearchNotEnabledCorrectly(userConfig, (message) => console.warn(message));
     registerKnownSecretsInRootKeychain(userConfig);
     return userConfig;
 }

src/common/search/vectorSearchEmbeddingsManager.ts

@@ -24,10 +24,8 @@ export type VectorFieldIndexDefinition = {
 export type VectorFieldValidationError = {
     path: string;
     expectedNumDimensions: number;
-    expectedQuantization: Quantization;
     actualNumDimensions: number | "unknown";
-    actualQuantization: Quantization | "unknown";
-    error: "dimension-mismatch" | "quantization-mismatch" | "not-a-vector" | "not-numeric";
+    error: "dimension-mismatch" | "not-a-vector" | "not-numeric";
 };
 
 export type EmbeddingNamespace = `${string}.${string}`;
@@ -116,9 +114,9 @@ export class VectorSearchEmbeddingsManager {
         if (embeddingValidationResults.length > 0) {
             const embeddingValidationMessages = embeddingValidationResults.map(
                 (validation) =>
-                    `- Field ${validation.path} is an embedding with ${validation.expectedNumDimensions} dimensions and ${validation.expectedQuantization}` +
-                    ` quantization, and the provided value is not compatible. Actual dimensions: ${validation.actualNumDimensions}, ` +
-                    `actual quantization: ${validation.actualQuantization}. Error: ${validation.error}`
+                    `- Field ${validation.path} is an embedding with ${validation.expectedNumDimensions} dimensions,` +
+                    ` and the provided value is not compatible. Actual dimensions: ${validation.actualNumDimensions},` +
+                    ` Error: ${validation.error}`
             );
 
             throw new MongoDBError(
@@ -179,16 +177,36 @@ export class VectorSearchEmbeddingsManager {
         let fieldRef: unknown = document;
 
         const constructError = (
-            details: Partial<Pick<VectorFieldValidationError, "error" | "actualNumDimensions" | "actualQuantization">>
+            details: Partial<Pick<VectorFieldValidationError, "error" | "actualNumDimensions">>
         ): VectorFieldValidationError => ({
             path: definition.path,
             expectedNumDimensions: definition.numDimensions,
-            expectedQuantization: definition.quantization,
             actualNumDimensions: details.actualNumDimensions ?? "unknown",
-            actualQuantization: details.actualQuantization ?? "unknown",
             error: details.error ?? "not-a-vector",
         });
 
+        const extractUnderlyingVector = (fieldRef: unknown): ArrayLike<unknown> | undefined => {
+            if (fieldRef instanceof BSON.Binary) {
+                try {
+                    return fieldRef.toFloat32Array();
+                } catch {
+                    // nothing to do here
+                }
+
+                try {
+                    return fieldRef.toBits();
+                } catch {
+                    // nothing to do here
+                }
+            }
+
+            if (Array.isArray(fieldRef)) {
+                return fieldRef as Array<unknown>;
+            }
+
+            return undefined;
+        };
+
         for (const field of fieldPath) {
             if (fieldRef && typeof fieldRef === "object" && field in fieldRef) {
                 fieldRef = (fieldRef as Record<string, unknown>)[field];
@@ -197,70 +215,25 @@ export class VectorSearchEmbeddingsManager {
             }
         }
 
-        switch (definition.quantization) {
-            // Because quantization is not defined by the user
-            // we have to trust them in the format they use.
-            case "none":
-                return undefined;
-            case "scalar":
-            case "binary":
-                if (fieldRef instanceof BSON.Binary) {
-                    try {
-                        const elements = fieldRef.toFloat32Array();
-                        if (elements.length !== definition.numDimensions) {
-                            return constructError({
-                                actualNumDimensions: elements.length,
-                                actualQuantization: "binary",
-                                error: "dimension-mismatch",
-                            });
-                        }
-
-                        return undefined;
-                    } catch {
-                        // bits are also supported
-                        try {
-                            const bits = fieldRef.toBits();
-                            if (bits.length !== definition.numDimensions) {
-                                return constructError({
-                                    actualNumDimensions: bits.length,
-                                    actualQuantization: "binary",
-                                    error: "dimension-mismatch",
-                                });
-                            }
-
-                            return undefined;
-                        } catch {
-                            return constructError({
-                                actualQuantization: "binary",
-                                error: "not-a-vector",
-                            });
-                        }
-                    }
-                } else {
-                    if (!Array.isArray(fieldRef)) {
-                        return constructError({
-                            error: "not-a-vector",
-                        });
-                    }
-
-                    if (fieldRef.length !== definition.numDimensions) {
-                        return constructError({
-                            actualNumDimensions: fieldRef.length,
-                            actualQuantization: "scalar",
-                            error: "dimension-mismatch",
-                        });
-                    }
-
-                    if (!fieldRef.every((e) => this.isANumber(e))) {
-                        return constructError({
-                            actualNumDimensions: fieldRef.length,
-                            actualQuantization: "scalar",
-                            error: "not-numeric",
-                        });
-                    }
-                }
+        const maybeVector = extractUnderlyingVector(fieldRef);
+        if (!maybeVector) {
+            return constructError({
+                error: "not-a-vector",
+            });
+        }
 
-                break;
+        if (maybeVector.length !== definition.numDimensions) {
+            return constructError({
+                actualNumDimensions: maybeVector.length,
+                error: "dimension-mismatch",
+            });
+        }
+
+        if (Array.isArray(maybeVector) && maybeVector.some((e) => !this.isANumber(e))) {
+            return constructError({
+                actualNumDimensions: maybeVector.length,
+                error: "not-numeric",
+            });
         }
 
         return undefined;

src/tools/mongodb/create/createIndex.ts

@@ -80,7 +80,7 @@ export class CreateIndexTool extends MongoDBToolBase {
                 ])
             )
             .describe(
-                "The index definition. Use 'classic' for standard indexes and 'vectorSearch' for vector search indexes"
+                `The index definition. Use 'classic' for standard indexes${this.isFeatureEnabled("vectorSearch") ? " and 'vectorSearch' for vector search indexes" : ""}.`
             ),
     };
 

src/tools/mongodb/metadata/explain.ts

@@ -4,7 +4,7 @@ import type { ToolArgs, OperationType } from "../../tool.js";
 import { formatUntrustedData } from "../../tool.js";
 import { z } from "zod";
 import type { Document } from "mongodb";
-import { AggregateArgs } from "../read/aggregate.js";
+import { getAggregateArgs } from "../read/aggregate.js";
 import { FindArgs } from "../read/find.js";
 import { CountArgs } from "../read/count.js";
 
@@ -20,7 +20,7 @@ export class ExplainTool extends MongoDBToolBase {
                 z.discriminatedUnion("name", [
                     z.object({
                         name: z.literal("aggregate"),
-                        arguments: z.object(AggregateArgs),
+                        arguments: z.object(getAggregateArgs(this.isFeatureEnabled("vectorSearch"))),
                     }),
                     z.object({
                         name: z.literal("find"),

src/tools/mongodb/mongodbSchemas.ts

@@ -42,7 +42,7 @@ export type EmbeddingParameters = {
 export const zSupportedEmbeddingParameters = zVoyageEmbeddingParameters.extend({ model: zVoyageModels });
 export type SupportedEmbeddingParameters = z.infer<typeof zSupportedEmbeddingParameters>;
 
-export const AnyVectorSearchStage = zEJSON();
+export const AnyAggregateStage = zEJSON();
 export const VectorSearchStage = z.object({
     $vectorSearch: z
         .object({

src/tools/mongodb/read/aggregate.ts

@@ -12,18 +12,17 @@ import { collectCursorUntilMaxBytesLimit } from "../../../helpers/collectCursorU
 import { operationWithFallback } from "../../../helpers/operationWithFallback.js";
 import { AGG_COUNT_MAX_TIME_MS_CAP, ONE_MB, CURSOR_LIMITS_TO_LLM_TEXT } from "../../../helpers/constants.js";
 import { LogId } from "../../../common/logger.js";
-import { AnyVectorSearchStage, VectorSearchStage } from "../mongodbSchemas.js";
+import { AnyAggregateStage, VectorSearchStage } from "../mongodbSchemas.js";
 import {
     assertVectorSearchFilterFieldsAreIndexed,
     type VectorSearchIndex,
 } from "../../../helpers/assertVectorSearchFilterFieldsAreIndexed.js";
 
-export const AggregateArgs = {
-    pipeline: z.array(z.union([AnyVectorSearchStage, VectorSearchStage])).describe(
-        `An array of aggregation stages to execute.  
+const pipelineDescriptionWithVectorSearch = `\
+An array of aggregation stages to execute.
 \`$vectorSearch\` **MUST** be the first stage of the pipeline, or the first stage of a \`$unionWith\` subpipeline.
 ### Usage Rules for \`$vectorSearch\`
-- **Unset embeddings:**  
+- **Unset embeddings:**
   Unless the user explicitly requests the embeddings, add an \`$unset\` stage **at the end of the pipeline** to remove the embedding field and avoid context limits. **The $unset stage in this situation is mandatory**.
 - **Pre-filtering:**
 If the user requests additional filtering, include filters in \`$vectorSearch.filter\` only for pre-filter fields in the vector index.
@@ -32,20 +31,28 @@ If the user requests additional filtering, include filters in \`$vectorSearch.fi
     For all remaining filters, add a $match stage after $vectorSearch.
 ### Note to LLM
 - If unsure which fields are filterable, use the collection-indexes tool to determine valid prefilter fields.
-- If no requested filters are valid prefilters, omit the filter key from $vectorSearch.`
-    ),
-    responseBytesLimit: z.number().optional().default(ONE_MB).describe(`\
+- If no requested filters are valid prefilters, omit the filter key from $vectorSearch.\
+`;
+
+const genericPipelineDescription = "An array of aggregation stages to execute.";
+
+export const getAggregateArgs = (vectorSearchEnabled: boolean) =>
+    ({
+        pipeline: z
+            .array(vectorSearchEnabled ? z.union([AnyAggregateStage, VectorSearchStage]) : AnyAggregateStage)
+            .describe(vectorSearchEnabled ? pipelineDescriptionWithVectorSearch : genericPipelineDescription),
+        responseBytesLimit: z.number().optional().default(ONE_MB).describe(`\
 The maximum number of bytes to return in the response. This value is capped by the server's configured maxBytesPerQuery and cannot be exceeded. \
 Note to LLM: If the entire aggregation result is required, use the "export" tool instead of increasing this limit.\
 `),
-};
+    }) as const;
 
 export class AggregateTool extends MongoDBToolBase {
     public name = "aggregate";
     protected description = "Run an aggregation against a MongoDB collection";
     protected argsShape = {
         ...DbOperationArgs,
-        ...AggregateArgs,
+        ...getAggregateArgs(this.isFeatureEnabled("vectorSearch")),
     };
     public operationType: OperationType = "read";
 

src/tools/mongodb/read/export.ts

@@ -6,7 +6,7 @@ import type { OperationType, ToolArgs } from "../../tool.js";
 import { DbOperationArgs, MongoDBToolBase } from "../mongodbTool.js";
 import { FindArgs } from "./find.js";
 import { jsonExportFormat } from "../../../common/exportsManager.js";
-import { AggregateArgs } from "./aggregate.js";
+import { getAggregateArgs } from "./aggregate.js";
 
 export class ExportTool extends MongoDBToolBase {
     public name = "export";
@@ -32,7 +32,9 @@ export class ExportTool extends MongoDBToolBase {
                         name: z
                             .literal("aggregate")
                             .describe("The literal name 'aggregate' to represent an aggregation cursor as target."),
-                        arguments: z.object(AggregateArgs).describe("The arguments for 'aggregate' operation."),
+                        arguments: z
+                            .object(getAggregateArgs(this.isFeatureEnabled("vectorSearch")))
+                            .describe("The arguments for 'aggregate' operation."),
                     }),
                 ])
             )

tests/integration/tools/mongodb/create/createIndex.test.ts

@@ -13,6 +13,22 @@ import { ObjectId, type Collection, type Document, type IndexDirection } from "m
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 
 describeWithMongoDB("createIndex tool when search is not enabled", (integration) => {
+    validateToolMetadata(integration, "create-index", "Create an index for a collection", [
+        ...databaseCollectionParameters,
+        {
+            name: "definition",
+            type: "array",
+            description: "The index definition. Use 'classic' for standard indexes.",
+            required: true,
+        },
+        {
+            name: "name",
+            type: "string",
+            description: "The name of the index",
+            required: false,
+        },
+    ]);
+
     it("doesn't allow creating vector search indexes", async () => {
         expect(integration.mcpServer().userConfig.previewFeatures).to.not.include("vectorSearch");
 
@@ -99,7 +115,7 @@ describeWithMongoDB(
                 name: "definition",
                 type: "array",
                 description:
-                    "The index definition. Use 'classic' for standard indexes and 'vectorSearch' for vector search indexes",
+                    "The index definition. Use 'classic' for standard indexes and 'vectorSearch' for vector search indexes.",
                 required: true,
             },
             {