@dalmia/calibrate-mcp 0.0.21 → 0.0.22

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: "0.0.22",
     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 0.0.22 2.915.1 0.1.0 @dalmia/calibrate-mcp"
   };
 });
 
@@ -53206,17 +53206,34 @@ 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$zodSchema.describe("- `agent`: built inside Calibrate\n- `connection`: your existing agent connected to Calibrate"),
+    updated_at: string2().describe("When the agent was last updated (ISO 8601 UTC)"),
+    uuid: string2().describe("ID of the agent")
+  });
+});
+
 // src/models/listagentsagentsgetop.ts
 var ListAgentsAgentsGetRequest$zodSchema, ListAgentsAgentsGetResponse$zodSchema;
 var init_listagentsagentsgetop = __esm(() => {
   init_zod();
+  init_agentsummary();
   init_httpvalidationerror();
-  init_routersagentsagentresponse();
   ListAgentsAgentsGetRequest$zodSchema = object({
     xAPIKey: string2().nullable().optional()
   });
   ListAgentsAgentsGetResponse$zodSchema = union([
-    array(RoutersAgentsAgentResponse$zodSchema).describe("Successful Response"),
+    array(AgentSummary$zodSchema).describe("Successful Response"),
     HTTPValidationError$zodSchema
   ]);
 });
@@ -54520,23 +54537,33 @@ 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/getagenttestsendpointagenttestsagentagentuuidtestsgetop.ts
@@ -54544,13 +54571,13 @@ var GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest$zodSchema, GetA
 var init_getagenttestsendpointagenttestsagentagentuuidtestsgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersagentteststestresponse();
+  init_testlistresponse();
   GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetRequest$zodSchema = object({
     agent_uuid: string2().describe("Agent whose linked tests to list"),
     xAPIKey: string2().nullable().optional()
   });
   GetAgentTestsEndpointAgentTestsAgentAgentUuidTestsGetResponse$zodSchema = union([
-    array(RoutersAgentTestsTestResponse$zodSchema).describe("Successful Response"),
+    array(TestListResponse$zodSchema).describe("Successful Response"),
     HTTPValidationError$zodSchema
   ]);
 });
@@ -55723,7 +55750,7 @@ 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"),
@@ -57461,11 +57488,24 @@ 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 +57527,7 @@ 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"),
@@ -58031,21 +58071,21 @@ 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 +58096,13 @@ 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 +58231,12 @@ var ListTestsTestsGetRequest$zodSchema, ListTestsTestsGetResponse$zodSchema;
 var init_listteststestsgetop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersteststestresponse();
+  init_testlistresponse();
   ListTestsTestsGetRequest$zodSchema = object({
     xAPIKey: string2().nullable().optional()
   });
   ListTestsTestsGetResponse$zodSchema = union([
-    array(RoutersTestsTestResponse$zodSchema).describe("Successful Response"),
+    array(TestListResponse$zodSchema).describe("Successful Response"),
     HTTPValidationError$zodSchema
   ]);
 });
@@ -58344,7 +58384,7 @@ var UpdateTestEndpointTestsTestUuidPutRequest$zodSchema, UpdateTestEndpointTests
 var init_updatetestendpointteststestuuidputop = __esm(() => {
   init_zod();
   init_httpvalidationerror();
-  init_routersteststestresponse();
+  init_testresponse();
   init_testupdate();
   UpdateTestEndpointTestsTestUuidPutRequest$zodSchema = object({
     body: TestUpdate$zodSchema,
@@ -58352,7 +58392,7 @@ var init_updatetestendpointteststestuuidputop = __esm(() => {
     xAPIKey: string2().nullable().optional()
   });
   UpdateTestEndpointTestsTestUuidPutResponse$zodSchema = union([
-    RoutersTestsTestResponse$zodSchema,
+    TestResponse$zodSchema,
     HTTPValidationError$zodSchema
   ]);
 });
@@ -58481,7 +58521,7 @@ 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: "0.0.22"
   });
   const getClient = deps.getSDK || (() => new CalibrateMcpCore({
     security: deps.security,
@@ -59687,7 +59727,7 @@ 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/v0.0.22/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 +62379,7 @@ var routes = buildRouteMap({
 var app = buildApplication(routes, {
   name: "mcp",
   versionInfo: {
-    currentVersion: "0.0.21"
+    currentVersion: "0.0.22"
   }
 });
 run(app, process4.argv.slice(2), buildContext(process4));
@@ -62347,5 +62387,5 @@ export {
   app
 };
 
-//# debugId=5A6E6EE0E3BB397164756E2164756E21
+//# debugId=5D121DD9360E345064756E2164756E21
 //# sourceMappingURL=mcp-server.js.map