@dalmia/calibrate-mcp 0.0.68 → 0.0.70

package/bin/mcp-server.js

@@ -1443,7 +1443,7 @@ var init_versions = __esm(() => {
   version2 = {
     major: 4,
     minor: 6,
-    patch: 6
+    patch: 7
   };
 });
 
@@ -52979,9 +52979,9 @@ var init_config = __esm(() => {
   SDK_METADATA = {
     language: "typescript",
     openapiDocVersion: "0.1.0",
-    sdkVersion: "0.0.68",
+    sdkVersion: "0.0.70",
     genVersion: "2.915.1",
-    userAgent: "speakeasy-sdk/mcp-typescript 0.0.68 2.915.1 0.1.0 @dalmia/calibrate-mcp"
+    userAgent: "speakeasy-sdk/mcp-typescript 0.0.70 2.915.1 0.1.0 @dalmia/calibrate-mcp"
   };
 });
 
@@ -57680,14 +57680,123 @@ var init_bulkcreateitemsresponse = __esm(() => {
   });
 });
 
+// src/models/itemmetadataentry.ts
+var ItemMetadataEntry$zodSchema;
+var init_itemmetadataentry = __esm(() => {
+  init_zod();
+  ItemMetadataEntry$zodSchema = object({
+    key: string2().describe("Name of the metadata entry"),
+    value: string2().describe("Value of the metadata entry")
+  }).describe("One key-value pair stored with an item and shown beside it on the labelling form");
+});
+
+// src/models/chatturn.ts
+var ChatTurn$zodSchema, ChatTurn$zodSchemaOutbound;
+var init_chatturn = __esm(() => {
+  init_zod();
+  init_schemas4();
+  ChatTurn$zodSchema = collectExtraKeys(object({
+    content: string2().nullable().optional().describe("Message text. Omit for turns that only carry tool calls"),
+    label: string2().nullable().optional().describe("Display name for the text in `content`, shown to annotators"),
+    role: string2().describe("Message author role in the conversation history"),
+    versions: record(string2(), string2()).nullable().optional().describe("Other forms of the same text, such as a translation or a human rewrite, keyed by the name annotators see for each. Shown to annotators only")
+  }).describe("One message in a conversation, in OpenAI chat format").catchall(any()), "AdditionalProperties", true);
+  ChatTurn$zodSchemaOutbound = object({
+    AdditionalProperties: record(string2(), any()).optional(),
+    content: nullable(string2()).optional(),
+    label: nullable(string2()).optional(),
+    role: string2(),
+    versions: nullable(record(string2(), string2())).optional()
+  }).describe("One message in a conversation, in OpenAI chat format").transform((v) => {
+    const { AdditionalProperties, ...rest } = v;
+    return {
+      ...rest,
+      ...AdditionalProperties
+    };
+  });
+});
+
+// src/models/toolcall.ts
+var ToolCall$zodSchema;
+var init_toolcall = __esm(() => {
+  init_zod();
+  ToolCall$zodSchema = object({
+    accept_any_arguments: boolean2().nullable().optional().describe("When `true`, only the tool name had to match and `arguments` was ignored. Carried over from a tool-call test's expected calls"),
+    arguments: record(string2(), any()).nullable().optional().describe("Argument values the agent passed to the tool. Omit when the call had none"),
+    output: any().optional().describe("What the tool returned for this call. Any JSON value. Omit when you do not record it"),
+    tool: string2().describe("Name of the tool the agent called")
+  }).describe("One call to a tool");
+});
+
+// src/models/turntext.ts
+var TurnText$zodSchema;
+var init_turntext = __esm(() => {
+  init_zod();
+  TurnText$zodSchema = object({
+    content: string2().describe("The text the LLM judge reads and annotators see by default"),
+    label: string2().nullable().optional().describe("Display name for the text in `content`, shown to annotators"),
+    versions: record(string2(), string2()).nullable().optional().describe("Other forms of the same text, such as a translation or a human rewrite, keyed by the name annotators see for each. Shown to annotators only")
+  }).describe("Text with a display name and other forms of it, sent in place of a plain string");
+});
+
+// src/models/itempayload.ts
+var AgentResponseUnion$zodSchema, ItemPayloadInput$zodSchema, Output$zodSchema, ItemPayload$zodSchema;
+var init_itempayload = __esm(() => {
+  init_zod();
+  init_chatturn();
+  init_toolcall();
+  init_turntext();
+  AgentResponseUnion$zodSchema = union([
+    TurnText$zodSchema,
+    string2()
+  ]).describe("The reply to judge. An empty string when the reply was a tool call. For `llm` tasks");
+  ItemPayloadInput$zodSchema = union([
+    TurnText$zodSchema,
+    string2()
+  ]).describe("The standalone input the agent was given. For `llm-general` tasks");
+  Output$zodSchema = union([
+    TurnText$zodSchema,
+    string2()
+  ]).describe("The output to judge. For `llm-general` tasks");
+  ItemPayload$zodSchema = object({
+    actual_tool_calls: array(ToolCall$zodSchema).nullable().optional().describe("Tool calls the agent made. Sending this marks the item as a tool call for a person to judge. For `llm` and `llm-general` tasks"),
+    agent_response: union([
+      TurnText$zodSchema,
+      string2()
+    ]).nullable().optional().describe("The reply to judge. An empty string when the reply was a tool call. For `llm` tasks"),
+    audio_path: string2().nullable().optional().describe("Storage key or `s3://` URI of the audio clip. For `tts` tasks"),
+    chat_history: array(ChatTurn$zodSchema).nullable().optional().describe("The conversation up to the reply being judged, ending at the user turn. For `llm` tasks"),
+    description: string2().nullable().optional().describe("A description of the item. For every task type"),
+    evaluator_variables: record(string2(), record(string2(), any())).nullable().optional().describe('Values for each evaluator\'s `{{"{{"}}variable}}` placeholders, keyed by evaluator ID. For `llm`, `llm-general` and `conversation` tasks'),
+    expected_tool_calls: array(ToolCall$zodSchema).nullable().optional().describe("Tool calls the agent should have made. Sending this marks the item as a tool call for a person to judge. For `llm` and `llm-general` tasks"),
+    input: union([
+      TurnText$zodSchema,
+      string2()
+    ]).nullable().optional().describe("The standalone input the agent was given. For `llm-general` tasks"),
+    name: string2().describe("Name of the item, unique within the task. For every task type"),
+    output: union([
+      TurnText$zodSchema,
+      string2()
+    ]).nullable().optional().describe("The output to judge. For `llm-general` tasks"),
+    predicted_transcript: string2().nullable().optional().describe("The transcript to judge. For `stt` tasks"),
+    reference_transcript: string2().nullable().optional().describe("The correct transcript. For `stt` tasks"),
+    text: string2().nullable().optional().describe("The text that was spoken. For `tts` tasks"),
+    tool_calls: array(ToolCall$zodSchema).nullable().optional().describe("Tool calls the agent made in place of a text reply. With no text reply, sending this marks the item as a tool call for a person to judge. For `llm` and `llm-general` tasks"),
+    transcript: array(ChatTurn$zodSchema).nullable().optional().describe("The full conversation to judge. For `conversation` tasks")
+  }).describe("What an annotation item holds. Which fields apply depends on the task type");
+});
+
 // src/models/annotationitempayload.ts
 var AnnotationItemPayload$zodSchema;
 var init_annotationitempayload = __esm(() => {
   init_zod();
+  init_itemmetadataentry();
+  init_itempayload();
   AnnotationItemPayload$zodSchema = object({
     annotations: record(string2(), any()).nullable().optional().describe("Human annotations to seed, keyed by evaluator ID. Each evaluator ID must be linked to the task. Put the judgement in `value`, a bool for binary or a number for rating, with optional `reasoning`. Requires `annotator_id`"),
     evaluator_results: record(string2(), any()).nullable().optional().describe("Evaluator scores to record, keyed by evaluator ID. Each evaluator ID must be linked to the task. Put the score in `value`, a bool for binary or a number within the scale for rating, with optional `reasoning` and `version_number`. Omit `version_number` to record against the evaluator's live version. An item recording a tool call takes the tool-call evaluator and no other, and every other item takes any evaluator except that one"),
-    payload: any().optional().describe("The item's payload. `name` is required and unique within the task for every type. The other fields depend on the task `type`:\n\n- `stt`: `name`, `reference_transcript`, `predicted_transcript`\n- `llm`: `name`, `chat_history` (list of `{role, content}` turns ending at the user turn), `agent_response` (the reply to judge), optional `evaluator_variables`\n- `llm-general`: `name`, `input`, `output`, optional `evaluator_variables`\n- `conversation`: `name`, `transcript` (list of `{role, content}` turns), optional `evaluator_variables`\n\n`evaluator_variables` maps an evaluator ID to that evaluator's `{{\"{{\"}}variable}}` values for this item")
+    metadata: array(ItemMetadataEntry$zodSchema).nullable().optional().describe("Key-value pairs stored with the item and shown beside it on the labelling form. Omit if you have none"),
+    payload: ItemPayload$zodSchema.describe("What an annotation item holds. Which fields apply depends on the task type")
   });
 });
 
@@ -59418,8 +59527,11 @@ must be in your workspace.
 var ItemUpdatePayload$zodSchema;
 var init_itemupdatepayload = __esm(() => {
   init_zod();
+  init_itemmetadataentry();
+  init_itempayload();
   ItemUpdatePayload$zodSchema = object({
-    payload: any().optional().describe("The item's new payload. `payload['name']` is still required and must be unique within the task"),
+    metadata: array(ItemMetadataEntry$zodSchema).nullable().optional().describe("Key-value pairs stored with the item and shown beside it on the labelling form. Omit to leave the stored pairs as they are. Send an empty list to clear them"),
+    payload: ItemPayload$zodSchema.describe("What an annotation item holds. Which fields apply depends on the task type"),
     uuid: string2().describe("ID of the item to update")
   });
 });
@@ -61610,13 +61722,13 @@ var init_traceturn = __esm(() => {
 });
 
 // src/models/traceingest.ts
-var Input$zodSchema, TraceIngest$zodSchema;
+var TraceIngestInput$zodSchema, TraceIngest$zodSchema;
 var init_traceingest = __esm(() => {
   init_zod();
   init_tracemetadataentry();
   init_traceoutput();
   init_traceturn();
-  Input$zodSchema = union([
+  TraceIngestInput$zodSchema = union([
     string2(),
     array(TraceTurn$zodSchema)
   ]).describe("What the agent was given for this turn. For a `general` agent, the standalone prompt as a string. For a `conversation` agent, the history up to the reported output, oldest turn first, in OpenAI chat format");
@@ -61783,7 +61895,7 @@ hits its trace limit.
 function createMCPServer(deps) {
   const server = new McpServer({
     name: "CalibrateMcp",
-    version: "0.0.68"
+    version: "0.0.70"
   });
   const getClient = deps.getSDK || (() => new CalibrateMcpCore({
     security: deps.security,
@@ -63072,7 +63184,7 @@ 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/v0.0.68/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/v0.0.70/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>
@@ -66026,7 +66138,7 @@ var routes = buildRouteMap({
 var app = buildApplication(routes, {
   name: "mcp",
   versionInfo: {
-    currentVersion: "0.0.68"
+    currentVersion: "0.0.70"
   }
 });
 run(app, process3.argv.slice(2), buildContext(process3));
@@ -66034,5 +66146,5 @@ export {
   app
 };
 
-//# debugId=EE93391F8D4C1CAE64756E2164756E21
+//# debugId=E1B5DB298336254064756E2164756E21
 //# sourceMappingURL=mcp-server.js.map