@dalmia/calibrate-mcp 0.0.21 → 0.0.23

package/bin/mcp-server.js

@@ -51392,9 +51392,9 @@ var init_config = __esm(() => {
   SDK_METADATA = {
     language: "typescript",
     openapiDocVersion: "0.1.0",
-    sdkVersion: "0.0.21",
+    sdkVersion: "0.0.23",
     genVersion: "2.915.1",
-    userAgent: "speakeasy-sdk/mcp-typescript 0.0.21 2.915.1 0.1.0 @dalmia/calibrate-mcp"
+    userAgent: "speakeasy-sdk/mcp-typescript 0.0.23 2.915.1 0.1.0 @dalmia/calibrate-mcp"
   };
 });
 
@@ -53206,17 +53206,50 @@ exist or belongs to another org.
   };
 });
 
+// src/models/agentsummary.ts
+var AgentSummaryType$zodSchema, AgentSummary$zodSchema;
+var init_agentsummary = __esm(() => {
+  init_zod();
+  AgentSummaryType$zodSchema = _enum([
+    "agent",
+    "connection"
+  ]).describe("- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate");
+  AgentSummary$zodSchema = object({
+    connection_verified: boolean2().nullable().optional().describe("Whether the agent's connection has been verified, for a `type=connection` agent"),
+    name: string2().describe("Name of the agent"),
+    type: AgentSummaryType$zodSchema.describe("- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate"),
+    updated_at: string2().describe("When the agent was last updated (ISO 8601 UTC)"),
+    uuid: string2().describe("ID of the agent")
+  });
+});
+
+// src/models/paginatedresponseagentsummary.ts
+var PaginatedResponseAgentSummary$zodSchema;
+var init_paginatedresponseagentsummary = __esm(() => {
+  init_zod();
+  init_agentsummary();
+  PaginatedResponseAgentSummary$zodSchema = object({
+    items: array(AgentSummary$zodSchema).describe("The page of results"),
+    limit: int().nullable().optional().describe("The applied `limit` (null when unbounded)"),
+    offset: int().default(0).describe("The applied `offset`"),
+    total: int().describe("Total number of items matching the query, before pagination")
+  });
+});
+
 // src/models/listagentsagentsgetop.ts
 var ListAgentsAgentsGetRequest$zodSchema, ListAgentsAgentsGetResponse$zodSchema;
 var init_listagentsagentsgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersagentsagentresponse();
+  init_paginatedresponseagentsummary();
   ListAgentsAgentsGetRequest$zodSchema = object({
+    limit: int().describe("Maximum number of items to return. Omit for no limit (all items)").nullable().optional(),
+    offset: int().default(0).describe("Number of items to skip before returning results"),
+    q: string2().describe("Case-insensitive substring search on `name`. Blank is a no-op").nullable().optional(),
     xAPIKey: string2().nullable().optional()
   });
   ListAgentsAgentsGetResponse$zodSchema = union([
-    array(RoutersAgentsAgentResponse$zodSchema).describe("Successful Response"),
+    PaginatedResponseAgentSummary$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -53233,6 +53266,11 @@ async function $do3(client$, request, options) {
   const payload$ = parsed$.value;
   const body$ = null;
   const path$ = pathToFunc("/agents")();
+  const query$ = encodeFormQuery({
+    limit: payload$?.limit,
+    offset: payload$?.offset,
+    q: payload$?.q
+  });
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
     "X-API-Key": encodeSimple("X-API-Key", payload$?.xAPIKey, {
@@ -53264,6 +53302,7 @@ async function $do3(client$, request, options) {
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
@@ -54520,22 +54559,45 @@ var init_agentTestsLink2 = __esm(() => {
   };
 });
 
-// src/models/routersagentteststestresponse.ts
-var RoutersAgentTestsTestResponseType$zodSchema, RoutersAgentTestsTestResponse$zodSchema;
-var init_routersagentteststestresponse = __esm(() => {
+// src/models/testlistconfig.ts
+var TestListConfig$zodSchema;
+var init_testlistconfig = __esm(() => {
   init_zod();
-  RoutersAgentTestsTestResponseType$zodSchema = _enum([
+  TestListConfig$zodSchema = object({
+    description: string2().nullable().optional().describe("Short description of the test, shown in list views and searched on")
+  });
+});
+
+// src/models/testlistresponse.ts
+var TestListResponseType$zodSchema, TestListResponse$zodSchema;
+var init_testlistresponse = __esm(() => {
+  init_zod();
+  init_testlistconfig();
+  TestListResponseType$zodSchema = _enum([
     "response",
     "tool_call",
     "conversation"
-  ]).describe("What the test checks:\n- `response`: judges the reply the agent generates\n- `tool_call`: checks the tool calls the agent makes\n- `conversation`: judges the full conversation after the agent replies");
-  RoutersAgentTestsTestResponse$zodSchema = object({
-    config: record(string2(), any()).nullable().optional().describe("Test configuration"),
+  ]).describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n");
+  TestListResponse$zodSchema = object({
+    config: TestListConfig$zodSchema.nullable().optional().describe("Trimmed config carrying only the test's description. Fetch the test by ID for the full config and evaluators"),
     created_at: string2().describe("When the test was created (ISO 8601 UTC)"),
     name: string2().describe("Name of the test"),
-    type: RoutersAgentTestsTestResponseType$zodSchema.describe("What the test checks:\n- `response`: judges the reply the agent generates\n- `tool_call`: checks the tool calls the agent makes\n- `conversation`: judges the full conversation after the agent replies"),
+    type: TestListResponseType$zodSchema.describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n"),
     updated_at: string2().describe("When the test was last updated (ISO 8601 UTC)"),
-    uuid: string2().describe("Test ID")
+    uuid: string2().describe("Unique ID for the test")
+  }).describe("Trimmed test shape for list/index endpoints. The full config and hydrated\nevaluators live on the detail endpoint (`GET /tests/{uuid}`); the list keeps\nonly what list and attach-dropdown views render.");
+});
+
+// src/models/paginatedresponsetestlistresponse.ts
+var PaginatedResponseTestListResponse$zodSchema;
+var init_paginatedresponsetestlistresponse = __esm(() => {
+  init_zod();
+  init_testlistresponse();
+  PaginatedResponseTestListResponse$zodSchema = object({
+    items: array(TestListResponse$zodSchema).describe("The page of results"),
+    limit: int().nullable().optional().describe("The applied `limit` (null when unbounded)"),
+    offset: int().default(0).describe("The applied `offset`"),
+    total: int().describe("Total number of items matching the query, before pagination")
   });
 });
 
@@ -54544,13 +54606,16 @@ var GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest$zodSchema, GetA
 var init_getagenttestsendpointagenttestsagentagentuuidtestsgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersagentteststestresponse();
+  init_paginatedresponsetestlistresponse();
   GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest$zodSchema = object({
     agent_uuid: string2().describe("Agent whose linked tests to list"),
+    limit: int().describe("Maximum number of items to return. Omit for no limit (all items)").nullable().optional(),
+    offset: int().default(0).describe("Number of items to skip before returning results"),
+    q: string2().describe("Case-insensitive substring search on `name`. Blank is a no-op").nullable().optional(),
     xAPIKey: string2().nullable().optional()
   });
   GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetResponse$zodSchema = union([
-    array(RoutersAgentTestsTestResponse$zodSchema).describe("Successful Response"),
+    PaginatedResponseTestListResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -54573,6 +54638,11 @@ async function $do11(client$, request, options) {
     })
   };
   const path$ = pathToFunc("/agent-tests/agent/{agent_uuid}/tests")(pathParams$);
+  const query$ = encodeFormQuery({
+    limit: payload$.limit,
+    offset: payload$.offset,
+    q: payload$.q
+  });
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
     "X-API-Key": encodeSimple("X-API-Key", payload$.xAPIKey, {
@@ -54604,6 +54674,7 @@ async function $do11(client$, request, options) {
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
@@ -54731,28 +54802,41 @@ var init_agenttestrunlistitem = __esm(() => {
   });
 });
 
-// src/models/agenttestrunsresponse.ts
-var AgentTestRunsResponse$zodSchema;
-var init_agenttestrunsresponse = __esm(() => {
+// src/models/paginatedresponseagenttestrunlistitem.ts
+var PaginatedResponseAgentTestRunListItem$zodSchema;
+var init_paginatedresponseagenttestrunlistitem = __esm(() => {
   init_zod();
   init_agenttestrunlistitem();
-  AgentTestRunsResponse$zodSchema = object({
-    runs: array(AgentTestRunListItem$zodSchema).describe("Test runs for the agent, most recently created first")
+  PaginatedResponseAgentTestRunListItem$zodSchema = object({
+    items: array(AgentTestRunListItem$zodSchema).describe("The page of results"),
+    limit: int().nullable().optional().describe("The applied `limit` (null when unbounded)"),
+    offset: int().default(0).describe("The applied `offset`"),
+    total: int().describe("Total number of items matching the query, before pagination")
   });
 });
 
 // src/models/getagenttestrunsagenttestsagentagentuuidrunsgetop.ts
-var GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetRequest$zodSchema, GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetResponse$zodSchema;
+var GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType$zodSchema, GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetRequest$zodSchema, GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetResponse$zodSchema;
 var init_getagenttestrunsagenttestsagentagentuuidrunsgetop = __esm(() => {
   init_zod();
-  init_agenttestrunsresponse();
   init_httpvalidationerror();
+  init_paginatedresponseagenttestrunlistitem();
+  init_taskstatus();
+  GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType$zodSchema = _enum([
+    "llm-unit-test",
+    "llm-benchmark"
+  ]).describe("Filter by run type. Omit to return both:\n- `llm-unit-test`: single runs of an agent's tests\n- `llm-benchmark`: multi-model comparisons");
   GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetRequest$zodSchema = object({
     agent_uuid: string2().describe("Agent whose test runs to list"),
+    has_failures: boolean2().describe("Filter by whether the run has any failing test case or model. `true` returns only runs with failures (or errors), `false` only clean runs. Omit for both").nullable().optional(),
+    limit: int().describe("Maximum number of items to return. Omit for no limit (all items)").nullable().optional(),
+    offset: int().default(0).describe("Number of items to skip before returning results"),
+    status: TaskStatus$zodSchema.nullable().optional().describe("Filter by run status. Omit for all statuses"),
+    type: GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetType$zodSchema.nullable().optional().describe("Filter by run type. Omit to return both:\n- `llm-unit-test`: single runs of an agent's tests\n- `llm-benchmark`: multi-model comparisons"),
     xAPIKey: string2().nullable().optional()
   });
   GetAgentTestRunsAgentTestsAgentAgentUuidRunsGetResponse$zodSchema = union([
-    AgentTestRunsResponse$zodSchema,
+    PaginatedResponseAgentTestRunListItem$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -54775,6 +54859,13 @@ async function $do12(client$, request, options) {
     })
   };
   const path$ = pathToFunc("/agent-tests/agent/{agent_uuid}/runs")(pathParams$);
+  const query$ = encodeFormQuery({
+    has_failures: payload$.has_failures,
+    limit: payload$.limit,
+    offset: payload$.offset,
+    status: payload$.status,
+    type: payload$.type
+  });
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
     "X-API-Key": encodeSimple("X-API-Key", payload$.xAPIKey, {
@@ -54806,6 +54897,7 @@ async function $do12(client$, request, options) {
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
@@ -55723,7 +55815,7 @@ var init_annotationtaskresponse = __esm(() => {
   AnnotationTaskResponse$zodSchema = object({
     created_at: string2().describe("When the task was created (ISO 8601 UTC)"),
     description: string2().nullable().optional().describe("The task's description"),
-    evaluators: array(record(string2(), any())).optional().describe("The evaluators linked to this task, in display order, each with its live version's rubric"),
+    evaluators: array(record(string2(), any())).optional().describe("The evaluators linked to this task, in display order. Fetching one task by ID enriches each with its live version's rubric. Listing tasks returns lightweight metadata (uuid, name, type, position) only"),
     item_count: int().nullable().optional().describe("Number of items in the task"),
     items: array(record(string2(), any())).optional().describe("The task's items, each with its agreement stats. You get these when you fetch one task by ID, not when you list tasks"),
     jobs: array(record(string2(), any())).optional().describe("The task's labelling jobs. You get these when you fetch one task by ID, not when you list tasks"),
@@ -56607,17 +56699,33 @@ var init_annotationTasksLinkEvaluator2 = __esm(() => {
   };
 });
 
+// src/models/paginatedresponseannotationtaskresponse.ts
+var PaginatedResponseAnnotationTaskResponse$zodSchema;
+var init_paginatedresponseannotationtaskresponse = __esm(() => {
+  init_zod();
+  init_annotationtaskresponse();
+  PaginatedResponseAnnotationTaskResponse$zodSchema = object({
+    items: array(AnnotationTaskResponse$zodSchema).describe("The page of results"),
+    limit: int().nullable().optional().describe("The applied `limit` (null when unbounded)"),
+    offset: int().default(0).describe("The applied `offset`"),
+    total: int().describe("Total number of items matching the query, before pagination")
+  });
+});
+
 // src/models/listannotationtasksannotationtasksgetop.ts
 var ListAnnotationTasksAnnotationTasksGetRequest$zodSchema, ListAnnotationTasksAnnotationTasksGetResponse$zodSchema;
 var init_listannotationtasksannotationtasksgetop = __esm(() => {
   init_zod();
-  init_annotationtaskresponse();
   init_httpvalidationerror();
+  init_paginatedresponseannotationtaskresponse();
   ListAnnotationTasksAnnotationTasksGetRequest$zodSchema = object({
+    limit: int().describe("Maximum number of items to return. Omit for no limit (all items)").nullable().optional(),
+    offset: int().default(0).describe("Number of items to skip before returning results"),
+    q: string2().describe("Case-insensitive substring search on `name`. Blank is a no-op").nullable().optional(),
     xAPIKey: string2().nullable().optional()
   });
   ListAnnotationTasksAnnotationTasksGetResponse$zodSchema = union([
-    array(AnnotationTaskResponse$zodSchema).describe("Successful Response"),
+    PaginatedResponseAnnotationTaskResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -56634,6 +56742,11 @@ async function $do23(client$, request, options) {
   const payload$ = parsed$.value;
   const body$ = null;
   const path$ = pathToFunc("/annotation-tasks")();
+  const query$ = encodeFormQuery({
+    limit: payload$?.limit,
+    offset: payload$?.offset,
+    q: payload$?.q
+  });
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
     "X-API-Key": encodeSimple("X-API-Key", payload$?.xAPIKey, {
@@ -56665,6 +56778,7 @@ async function $do23(client$, request, options) {
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
@@ -57461,11 +57575,24 @@ var init_evaluatorsGet2 = __esm(() => {
   };
 });
 
+// src/models/evaluatorliveversionsummary.ts
+var EvaluatorLiveVersionSummary$zodSchema;
+var init_evaluatorliveversionsummary = __esm(() => {
+  init_zod();
+  init_variablespec();
+  EvaluatorLiveVersionSummary$zodSchema = object({
+    judge_model: string2().describe("The model that runs the judge, named the way its provider does, for example `openai/gpt-4.1` or `anthropic/claude-sonnet-4`"),
+    uuid: string2().describe("Version ID"),
+    variables: array(VariableSpec$zodSchema).nullable().optional().describe("Declared prompt variables"),
+    version_number: int().describe("The version's number. The first version is 1, and it goes up by one for each new version of the evaluator")
+  }).describe("Slim view of the live version for list results, carrying only what the list needs.");
+});
+
 // src/models/routersevaluatorsevaluatorresponse.ts
 var RoutersEvaluatorsEvaluatorResponseEvaluatorType$zodSchema, RoutersEvaluatorsEvaluatorResponseDataType$zodSchema, RoutersEvaluatorsEvaluatorResponseOutputType$zodSchema, RoutersEvaluatorsEvaluatorResponse$zodSchema;
 var init_routersevaluatorsevaluatorresponse = __esm(() => {
   init_zod();
-  init_evaluatorversionresponse();
+  init_evaluatorliveversionsummary();
   RoutersEvaluatorsEvaluatorResponseEvaluatorType$zodSchema = _enum([
     "tts",
     "stt",
@@ -57487,7 +57614,7 @@ var init_routersevaluatorsevaluatorresponse = __esm(() => {
     description: string2().nullable().optional().describe("What the evaluator checks"),
     evaluator_type: RoutersEvaluatorsEvaluatorResponseEvaluatorType$zodSchema.describe("What the evaluator judges:\n\n- `tts`: TTS audio\n- `stt`: one transcript\n- `llm`: a reply with its conversation history\n- `llm-general`: a standalone input and output pair\n- `conversation`: a full conversation\n"),
     is_default: boolean2().describe("True for a built-in default evaluator, which you can't edit. False for an evaluator you created, which you can edit and add versions to"),
-    live_version: EvaluatorVersionResponse$zodSchema.nullable().optional().describe("The version that is currently live"),
+    live_version: EvaluatorLiveVersionSummary$zodSchema.nullable().optional().describe("The version that is currently live"),
     live_version_id: string2().nullable().optional().describe("ID of the version that is currently live"),
     name: string2().describe("Evaluator name"),
     output_type: RoutersEvaluatorsEvaluatorResponseOutputType$zodSchema.describe("How the evaluator scores:\n\n- `binary`: pass or fail\n- `rating`: a numeric score, using the scale in `output_config`\n"),
@@ -57497,12 +57624,25 @@ var init_routersevaluatorsevaluatorresponse = __esm(() => {
   });
 });
 
+// src/models/paginatedresponseevaluatorresponse.ts
+var PaginatedResponseEvaluatorResponse$zodSchema;
+var init_paginatedresponseevaluatorresponse = __esm(() => {
+  init_zod();
+  init_routersevaluatorsevaluatorresponse();
+  PaginatedResponseEvaluatorResponse$zodSchema = object({
+    items: array(RoutersEvaluatorsEvaluatorResponse$zodSchema).describe("The page of results"),
+    limit: int().nullable().optional().describe("The applied `limit` (null when unbounded)"),
+    offset: int().default(0).describe("The applied `offset`"),
+    total: int().describe("Total number of items matching the query, before pagination")
+  });
+});
+
 // src/models/listevaluatorsevaluatorsgetop.ts
 var ListEvaluatorsEvaluatorsGetEvaluatorType$zodSchema, ListEvaluatorsEvaluatorsGetDataType$zodSchema, ListEvaluatorsEvaluatorsGetRequest$zodSchema, ListEvaluatorsEvaluatorsGetResponse$zodSchema;
 var init_listevaluatorsevaluatorsgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersevaluatorsevaluatorresponse();
+  init_paginatedresponseevaluatorresponse();
   ListEvaluatorsEvaluatorsGetEvaluatorType$zodSchema = _enum([
     "tts",
     "stt",
@@ -57518,10 +57658,13 @@ var init_listevaluatorsevaluatorsgetop = __esm(() => {
     data_type: ListEvaluatorsEvaluatorsGetDataType$zodSchema.nullable().optional().describe("Filter by modality. Omit for all"),
     evaluator_type: ListEvaluatorsEvaluatorsGetEvaluatorType$zodSchema.nullable().optional().describe("Filter by what the evaluator judges. Omit for all types"),
     include_defaults: boolean2().default(true).describe("When `true`, include the built-in default evaluators alongside the ones you created"),
+    limit: int().describe("Maximum number of items to return. Omit for no limit (all items)").nullable().optional(),
+    offset: int().default(0).describe("Number of items to skip before returning results"),
+    q: string2().describe("Case-insensitive substring search on `name`. Blank is a no-op").nullable().optional(),
     xAPIKey: string2().nullable().optional()
   });
   ListEvaluatorsEvaluatorsGetResponse$zodSchema = union([
-    array(RoutersEvaluatorsEvaluatorResponse$zodSchema).describe("Successful Response"),
+    PaginatedResponseEvaluatorResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -57541,7 +57684,10 @@ async function $do28(client$, request, options) {
   const query$ = encodeFormQuery({
     data_type: payload$?.data_type,
     evaluator_type: payload$?.evaluator_type,
-    include_defaults: payload$?.include_defaults
+    include_defaults: payload$?.include_defaults,
+    limit: payload$?.limit,
+    offset: payload$?.offset,
+    q: payload$?.q
   });
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
@@ -58031,21 +58177,21 @@ Requires a Calibrate API key (CALIBRATE_API_KEY).
   };
 });
 
-// src/models/routersteststestresponse.ts
-var RoutersTestsTestResponseType$zodSchema, RoutersTestsTestResponse$zodSchema;
-var init_routersteststestresponse = __esm(() => {
+// src/models/testresponse.ts
+var TestResponseType$zodSchema, TestResponse$zodSchema;
+var init_testresponse = __esm(() => {
   init_zod();
-  RoutersTestsTestResponseType$zodSchema = _enum([
+  TestResponseType$zodSchema = _enum([
     "response",
     "tool_call",
     "conversation"
   ]).describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n");
-  RoutersTestsTestResponse$zodSchema = object({
+  TestResponse$zodSchema = object({
     config: record(string2(), any()).nullable().optional().describe("The stored config: `history`, `evaluation`, and an optional `settings`"),
     created_at: string2().describe("When the test was created (ISO 8601 UTC)"),
     evaluators: array(record(string2(), any())).optional().describe("Linked evaluators, resolved to their current live version at read time"),
     name: string2().describe("Name of the test"),
-    type: RoutersTestsTestResponseType$zodSchema.describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n"),
+    type: TestResponseType$zodSchema.describe("What the test judges:\n\n- `response`: judges the generated reply\n- `tool_call`: diffs the generated tool calls\n- `conversation`: judges the full conversation\n"),
     updated_at: string2().describe("When the test was last updated (ISO 8601 UTC)"),
     uuid: string2().describe("Unique ID for the test")
   });
@@ -58056,13 +58202,13 @@ var GetTestEndpointTestsTestUuidGetRequest$zodSchema, GetTestEndpointTestsTestUu
 var init_gettestendpointteststestuuidgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersteststestresponse();
+  init_testresponse();
   GetTestEndpointTestsTestUuidGetRequest$zodSchema = object({
     test_uuid: string2().describe("Test to retrieve"),
     xAPIKey: string2().nullable().optional()
   });
   GetTestEndpointTestsTestUuidGetResponse$zodSchema = union([
-    RoutersTestsTestResponse$zodSchema,
+    TestResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -58191,12 +58337,15 @@ var ListTestsTestsGetRequest$zodSchema, ListTestsTestsGetResponse$zodSchema;
 var init_listteststestsgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersteststestresponse();
+  init_paginatedresponsetestlistresponse();
   ListTestsTestsGetRequest$zodSchema = object({
+    limit: int().describe("Maximum number of items to return. Omit for no limit (all items)").nullable().optional(),
+    offset: int().default(0).describe("Number of items to skip before returning results"),
+    q: string2().describe("Case-insensitive substring search on `name`. Blank is a no-op").nullable().optional(),
     xAPIKey: string2().nullable().optional()
   });
   ListTestsTestsGetResponse$zodSchema = union([
-    array(RoutersTestsTestResponse$zodSchema).describe("Successful Response"),
+    PaginatedResponseTestListResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -58213,6 +58362,11 @@ async function $do32(client$, request, options) {
   const payload$ = parsed$.value;
   const body$ = null;
   const path$ = pathToFunc("/tests")();
+  const query$ = encodeFormQuery({
+    limit: payload$?.limit,
+    offset: payload$?.offset,
+    q: payload$?.q
+  });
   const headers$ = new Headers(compactMap({
     Accept: "application/json",
     "X-API-Key": encodeSimple("X-API-Key", payload$?.xAPIKey, {
@@ -58244,6 +58398,7 @@ async function $do32(client$, request, options) {
     baseURL: options?.serverURL,
     path: path$,
     headers: headers$,
+    query: query$,
     body: body$,
     userAgent: client$._options.userAgent,
     timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
@@ -58344,7 +58499,7 @@ var UpdateTestEndpointTestsTestUuidPutRequest$zodSchema, UpdateTestEndpointTests
 var init_updatetestendpointteststestuuidputop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersteststestresponse();
+  init_testresponse();
   init_testupdate();
   UpdateTestEndpointTestsTestUuidPutRequest$zodSchema = object({
     body: TestUpdate$zodSchema,
@@ -58352,7 +58507,7 @@ var init_updatetestendpointteststestuuidputop = __esm(() => {
     xAPIKey: string2().nullable().optional()
   });
   UpdateTestEndpointTestsTestUuidPutResponse$zodSchema = union([
-    RoutersTestsTestResponse$zodSchema,
+    TestResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -58481,7 +58636,7 @@ fields to change (name, config). Returns the updated test. Fails with
 function createMCPServer(deps) {
   const server = new McpServer({
     name: "CalibrateMcp",
-    version: "0.0.21"
+    version: "0.0.23"
   });
   const getClient = deps.getSDK || (() => new CalibrateMcpCore({
     security: deps.security,
@@ -59687,7 +59842,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
         <h1>Instructions</h1>
         <p>One-click installation for Claude Desktop users</p>
         <div class="instruction-item">
-          <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.21/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
+          <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.23/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
             \uD83D\uDCE5 Download MCP Bundle
           </a>
         </div>
@@ -62339,7 +62494,7 @@ var routes = buildRouteMap({
 var app = buildApplication(routes, {
   name: "mcp",
   versionInfo: {
-    currentVersion: "0.0.21"
+    currentVersion: "0.0.23"
   }
 });
 run(app, process4.argv.slice(2), buildContext(process4));
@@ -62347,5 +62502,5 @@ export {
   app
 };
 
-//# debugId=5A6E6EE0E3BB397164756E2164756E21
+//# debugId=C650F485386E271664756E2164756E21
 //# sourceMappingURL=mcp-server.js.map