chore: more accuracy test tweaks

nirinchev · nirinchev · commit fc9325c6564b · 2025-10-22T14:29:46.000+02:00
diff --git a/src/tools/mongodb/metadata/collectionSchema.ts b/src/tools/mongodb/metadata/collectionSchema.ts
@@ -21,7 +21,7 @@ export class CollectionSchemaTool extends MongoDBToolBase {
             .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.`
+                `The maximum number of bytes to return in the response. This value is capped by the server's configured maxBytesPerQuery and cannot be exceeded.`
             ),
     };
 
diff --git a/src/tools/mongodb/read/aggregate.ts b/src/tools/mongodb/read/aggregate.ts
@@ -17,7 +17,7 @@ import { LogId } from "../../../common/logger.js";
 export const AggregateArgs = {
     pipeline: z.array(zEJSON()).describe("An array of aggregation stages to execute"),
     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. \
+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.\
 `),
 };
diff --git a/src/tools/mongodb/read/find.ts b/src/tools/mongodb/read/find.ts
@@ -30,7 +30,7 @@ export const FindArgs = {
             "A document, describing the sort order, matching the syntax of the sort argument of cursor.sort(). The keys of the object are the fields to sort on, while the values are the sort directions (1 for ascending, -1 for descending)."
         ),
     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. \
+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 query result is required, use the "export" tool instead of increasing this limit.\
 `),
 };
diff --git a/tests/accuracy/collectionSchema.test.ts b/tests/accuracy/collectionSchema.test.ts
@@ -1,26 +1,41 @@
 import { describeAccuracyTests } from "./sdk/describeAccuracyTests.js";
+import { Matcher } from "./sdk/matcher.js";
+
+const listCollectionsOptionalCall = {
+    toolName: "list-collections",
+    parameters: {
+        database: "mflix",
+    },
+    optional: true,
+};
 
 describeAccuracyTests([
     {
         prompt: "Is there a title field in 'mflix.movies' namespace?",
         expectedToolCalls: [
+            listCollectionsOptionalCall,
             {
                 toolName: "collection-schema",
                 parameters: {
                     database: "mflix",
                     collection: "movies",
+                    sampleSize: Matcher.anyOf(Matcher.undefined, Matcher.number()),
+                    responseBytesLimit: Matcher.anyOf(Matcher.undefined, Matcher.number()),
                 },
             },
         ],
     },
     {
         prompt: "What is the type of value stored in title field in movies collection in mflix database?",
         expectedToolCalls: [
+            listCollectionsOptionalCall,
             {
                 toolName: "collection-schema",
                 parameters: {
                     database: "mflix",
                     collection: "movies",
+                    sampleSize: Matcher.anyOf(Matcher.undefined, Matcher.number()),
+                    responseBytesLimit: Matcher.anyOf(Matcher.undefined, Matcher.number()),
                 },
             },
         ],
diff --git a/tests/accuracy/explain.test.ts b/tests/accuracy/explain.test.ts
@@ -45,6 +45,7 @@ describeAccuracyTests([
                                         $match: { release_year: 2020 },
                                     },
                                 ],
+                                responseBytesLimit: Matcher.anyOf(Matcher.undefined, Matcher.number()),
                             },
                         },
                     ],
diff --git a/tests/accuracy/export.test.ts b/tests/accuracy/export.test.ts
@@ -14,7 +14,9 @@ describeAccuracyTests([
                     exportTarget: [
                         {
                             name: "find",
-                            arguments: {},
+                            arguments: {
+                                filter: Matcher.emptyObjectOrUndefined,
+                            },
                         },
                     ],
                     jsonExportFormat: Matcher.anyValue,
diff --git a/tests/accuracy/find.test.ts b/tests/accuracy/find.test.ts
@@ -1,10 +1,27 @@
+import type { ExpectedToolCall } from "./sdk/accuracyResultStorage/resultStorage.js";
 import { describeAccuracyTests } from "./sdk/describeAccuracyTests.js";
 import { Matcher } from "./sdk/matcher.js";
 
+const optionalListCalls: (database: string) => ExpectedToolCall[] = (database) => [
+    {
+        toolName: "list-databases",
+        parameters: {},
+        optional: true,
+    },
+    {
+        toolName: "list-collections",
+        parameters: {
+            database,
+        },
+        optional: true,
+    },
+];
+
 describeAccuracyTests([
     {
         prompt: "List all the movies in 'mflix.movies' namespace.",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
@@ -18,6 +35,7 @@ describeAccuracyTests([
     {
         prompt: "List all the documents in 'comics.books' namespace.",
         expectedToolCalls: [
+            ...optionalListCalls("comics"),
             {
                 toolName: "find",
                 parameters: {
@@ -31,6 +49,7 @@ describeAccuracyTests([
     {
         prompt: "Find all the movies in 'mflix.movies' namespace with runtime less than 100.",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
@@ -46,6 +65,7 @@ describeAccuracyTests([
     {
         prompt: "Find all movies in 'mflix.movies' collection where director is 'Christina Collins'",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
@@ -61,6 +81,7 @@ describeAccuracyTests([
     {
         prompt: "Give me all the movie titles available in 'mflix.movies' namespace",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
@@ -81,6 +102,7 @@ describeAccuracyTests([
     {
         prompt: "Use 'mflix.movies' namespace to answer who were casted in the movie 'Certain Fish'",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
@@ -99,6 +121,7 @@ describeAccuracyTests([
     {
         prompt: "From the mflix.movies namespace, give me first 2 movies of Horror genre sorted ascending by their runtime",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
@@ -112,8 +135,9 @@ describeAccuracyTests([
         ],
     },
     {
-        prompt: "I want a COMPLETE list of all the movies ONLY from 'mflix.movies' namespace.",
+        prompt: "I want an exported COMPLETE list of all the movies ONLY from 'mflix.movies' namespace.",
         expectedToolCalls: [
+            ...optionalListCalls("mflix"),
             {
                 toolName: "find",
                 parameters: {
diff --git a/tests/accuracy/getPerformanceAdvisor.test.ts b/tests/accuracy/getPerformanceAdvisor.test.ts
@@ -1,5 +1,6 @@
 import { describeAccuracyTests } from "./sdk/describeAccuracyTests.js";
 import type { CallToolResult } from "@modelcontextprotocol/sdk/types.js";
+import { Matcher } from "./sdk/matcher.js";
 
 // Shared mock tool implementations
 const mockedTools = {
@@ -127,6 +128,15 @@ describeAccuracyTests([
                 parameters: {
                     projectId: "mflix",
                     clusterName: "mflix-cluster",
+                    operations: Matcher.anyOf(
+                        Matcher.undefined,
+                        Matcher.value([
+                            "suggestedIndexes",
+                            "dropIndexSuggestions",
+                            "slowQueryLogs",
+                            "schemaSuggestions",
+                        ])
+                    ),
                 },
             },
         ],
diff --git a/tests/accuracy/logs.test.ts b/tests/accuracy/logs.test.ts
@@ -9,6 +9,7 @@ describeAccuracyTests([
                 toolName: "mongodb-logs",
                 parameters: {
                     type: "startupWarnings",
+                    limit: Matcher.anyOf(Matcher.undefined, Matcher.number()),
                 },
             },
         ],
diff --git a/tests/integration/tools/mongodb/metadata/collectionSchema.test.ts b/tests/integration/tools/mongodb/metadata/collectionSchema.test.ts
@@ -26,7 +26,7 @@ describeWithMongoDB("collectionSchema tool", (integration) => {
         {
             name: "responseBytesLimit",
             type: "number",
-            description: `The maximum number of bytes to return in the response. This value is capped by the server’s configured maxBytesPerQuery and cannot be exceeded.`,
+            description: `The maximum number of bytes to return in the response. This value is capped by the server's configured maxBytesPerQuery and cannot be exceeded.`,
             required: false,
         },
     ]);
diff --git a/tests/integration/tools/mongodb/read/aggregate.test.ts b/tests/integration/tools/mongodb/read/aggregate.test.ts
@@ -26,8 +26,7 @@ describeWithMongoDB("aggregate tool", (integration) => {
         },
         {
             name: "responseBytesLimit",
-            description:
-                '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.',
+            description: `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.`,
             type: "number",
             required: false,
         },
diff --git a/tests/integration/tools/mongodb/read/find.test.ts b/tests/integration/tools/mongodb/read/find.test.ts
@@ -56,8 +56,7 @@ describeWithMongoDB("find tool with default configuration", (integration) => {
         },
         {
             name: "responseBytesLimit",
-            description:
-                '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 query result is required, use the "export" tool instead of increasing this limit.',
+            description: `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 query result is required, use the "export" tool instead of increasing this limit.`,
             type: "number",
             required: false,
         },

Original file line number	Diff line number	Diff line change
`@@ -1,10 +1,27 @@`
	`1`	`+import type { ExpectedToolCall } from "./sdk/accuracyResultStorage/resultStorage.js";`
`1`	`2`	`import { describeAccuracyTests } from "./sdk/describeAccuracyTests.js";`
`2`	`3`	`import { Matcher } from "./sdk/matcher.js";`
`3`	`4`
	`5`	`+const optionalListCalls: (database: string) => ExpectedToolCall[] = (database) => [`
	`6`	`+ {`
	`7`	`+ toolName: "list-databases",`
	`8`	`+ parameters: {},`
	`9`	`+ optional: true,`
	`10`	`+ },`
	`11`	`+ {`
	`12`	`+ toolName: "list-collections",`
	`13`	`+ parameters: {`
	`14`	`+ database,`
	`15`	`+ },`
	`16`	`+ optional: true,`
	`17`	`+ },`
	`18`	`+];`
	`19`	`+`
`4`	`20`	`describeAccuracyTests([`
`5`	`21`	`{`
`6`	`22`	`prompt: "List all the movies in 'mflix.movies' namespace.",`
`7`	`23`	`expectedToolCalls: [`
	`24`	`+ ...optionalListCalls("mflix"),`
`8`	`25`	`{`
`9`	`26`	`toolName: "find",`
`10`	`27`	`parameters: {`
`@@ -18,6 +35,7 @@ describeAccuracyTests([`
`18`	`35`	`{`
`19`	`36`	`prompt: "List all the documents in 'comics.books' namespace.",`
`20`	`37`	`expectedToolCalls: [`
	`38`	`+ ...optionalListCalls("comics"),`
`21`	`39`	`{`
`22`	`40`	`toolName: "find",`
`23`	`41`	`parameters: {`
`@@ -31,6 +49,7 @@ describeAccuracyTests([`
`31`	`49`	`{`
`32`	`50`	`prompt: "Find all the movies in 'mflix.movies' namespace with runtime less than 100.",`
`33`	`51`	`expectedToolCalls: [`
	`52`	`+ ...optionalListCalls("mflix"),`
`34`	`53`	`{`
`35`	`54`	`toolName: "find",`
`36`	`55`	`parameters: {`
`@@ -46,6 +65,7 @@ describeAccuracyTests([`
`46`	`65`	`{`
`47`	`66`	`prompt: "Find all movies in 'mflix.movies' collection where director is 'Christina Collins'",`
`48`	`67`	`expectedToolCalls: [`
	`68`	`+ ...optionalListCalls("mflix"),`
`49`	`69`	`{`
`50`	`70`	`toolName: "find",`
`51`	`71`	`parameters: {`
`@@ -61,6 +81,7 @@ describeAccuracyTests([`
`61`	`81`	`{`
`62`	`82`	`prompt: "Give me all the movie titles available in 'mflix.movies' namespace",`
`63`	`83`	`expectedToolCalls: [`
	`84`	`+ ...optionalListCalls("mflix"),`
`64`	`85`	`{`
`65`	`86`	`toolName: "find",`
`66`	`87`	`parameters: {`
`@@ -81,6 +102,7 @@ describeAccuracyTests([`
`81`	`102`	`{`
`82`	`103`	`prompt: "Use 'mflix.movies' namespace to answer who were casted in the movie 'Certain Fish'",`
`83`	`104`	`expectedToolCalls: [`
	`105`	`+ ...optionalListCalls("mflix"),`
`84`	`106`	`{`
`85`	`107`	`toolName: "find",`
`86`	`108`	`parameters: {`
`@@ -99,6 +121,7 @@ describeAccuracyTests([`
`99`	`121`	`{`
`100`	`122`	`prompt: "From the mflix.movies namespace, give me first 2 movies of Horror genre sorted ascending by their runtime",`
`101`	`123`	`expectedToolCalls: [`
	`124`	`+ ...optionalListCalls("mflix"),`
`102`	`125`	`{`
`103`	`126`	`toolName: "find",`
`104`	`127`	`parameters: {`
`@@ -112,8 +135,9 @@ describeAccuracyTests([`
`112`	`135`	`],`
`113`	`136`	`},`
`114`	`137`	`{`
`115`		`- prompt: "I want a COMPLETE list of all the movies ONLY from 'mflix.movies' namespace.",`
	`138`	`+ prompt: "I want an exported COMPLETE list of all the movies ONLY from 'mflix.movies' namespace.",`
`116`	`139`	`expectedToolCalls: [`
	`140`	`+ ...optionalListCalls("mflix"),`
`117`	`141`	`{`
`118`	`142`	`toolName: "find",`
`119`	`143`	`parameters: {`
Original file line number	Diff line number	Diff line change
`@@ -26,7 +26,7 @@ describeWithMongoDB("collectionSchema tool", (integration) => {`
`26`	`26`	`{`
`27`	`27`	`name: "responseBytesLimit",`
`28`	`28`	`type: "number",`
`29`		- description: `The maximum number of bytes to return in the response. This value is capped by the server’s configured maxBytesPerQuery and cannot be exceeded.`,
	`29`	+ description: `The maximum number of bytes to return in the response. This value is capped by the server's configured maxBytesPerQuery and cannot be exceeded.`,
`30`	`30`	`required: false,`
`31`	`31`	`},`
`32`	`32`	`]);`