@dalmia/calibrate-mcp 0.0.59 → 0.0.60

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -915,7 +915,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
915
915
  <h1>Instructions</h1>
916
916
  <p>One-click installation for Claude Desktop users</p>
917
917
  <div class="instruction-item">
918
- <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.59/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
918
+ <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.60/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
919
919
  📥 Download MCP Bundle
920
920
  </a>
921
921
  </div>
@@ -35,8 +35,8 @@ export declare function serverURLFromOptions(options: SDKOptions): URL | null;
35
35
  export declare const SDK_METADATA: {
36
36
  readonly language: "typescript";
37
37
  readonly openapiDocVersion: "0.1.0";
38
- readonly sdkVersion: "0.0.59";
38
+ readonly sdkVersion: "0.0.60";
39
39
  readonly genVersion: "2.915.1";
40
- readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.59 2.915.1 0.1.0 @dalmia/calibrate-mcp";
40
+ readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.60 2.915.1 0.1.0 @dalmia/calibrate-mcp";
41
41
  };
42
42
  //# sourceMappingURL=config.d.ts.map
package/esm/lib/config.js CHANGED
@@ -27,8 +27,8 @@ export function serverURLFromOptions(options) {
27
27
  export const SDK_METADATA = {
28
28
  language: "typescript",
29
29
  openapiDocVersion: "0.1.0",
30
- sdkVersion: "0.0.59",
30
+ sdkVersion: "0.0.60",
31
31
  genVersion: "2.915.1",
32
- userAgent: "speakeasy-sdk/mcp-typescript 0.0.59 2.915.1 0.1.0 @dalmia/calibrate-mcp",
32
+ userAgent: "speakeasy-sdk/mcp-typescript 0.0.60 2.915.1 0.1.0 @dalmia/calibrate-mcp",
33
33
  };
34
34
  //# sourceMappingURL=config.js.map
@@ -18,7 +18,7 @@ const routes = buildRouteMap({
18
18
  export const app = buildApplication(routes, {
19
19
  name: "mcp",
20
20
  versionInfo: {
21
- currentVersion: "0.0.59",
21
+ currentVersion: "0.0.60",
22
22
  },
23
23
  });
24
24
  run(app, process.argv.slice(2), buildContext(process));
@@ -51,7 +51,7 @@ import { tool$tracesCreate } from "./tools/tracesCreate.js";
51
51
  export function createMCPServer(deps) {
52
52
  const server = new McpServer({
53
53
  name: "CalibrateMcp",
54
- version: "0.0.59",
54
+ version: "0.0.60",
55
55
  });
56
56
  const getClient = deps.getSDK || (() => new CalibrateMcpCore({
57
57
  security: deps.security,
@@ -4,7 +4,7 @@
4
4
  import * as z from "zod";
5
5
  export const BenchmarkRequest$zodSchema = z.object({
6
6
  models: z.array(z.string()).describe("Model names to benchmark"),
7
- parallel_models: z.boolean().default(true).describe("Whether to run the models at the same time. Set false to run them one after another"),
7
+ parallel_models: z.boolean().default(true).describe("How to run the models. `true` runs several at a time, `false` runs each one only after the one before it has finished. Use `false` to keep the load on your own agent down"),
8
8
  test_uuids: z.array(z.string()).nullable().optional().describe("A subset of the agent's linked tests to benchmark. Each ID must be linked to the agent. Omit to run all linked tests"),
9
9
  });
10
10
  //# sourceMappingURL=benchmarkrequest.js.map
@@ -1 +1 @@
1
- {"version":3,"file":"benchmarkrequest.js","sourceRoot":"","sources":["../../src/models/benchmarkrequest.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAQzB,MAAM,CAAC,MAAM,0BAA0B,GAAgC,CAAC,CAAC,MAAM,CAC7E;IACE,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,CAAC,0BAA0B,CAAC;IAChE,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,QAAQ,CACjD,qFAAqF,CACtF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,sHAAsH,CACvH;CACF,CACF,CAAC"}
1
+ {"version":3,"file":"benchmarkrequest.js","sourceRoot":"","sources":["../../src/models/benchmarkrequest.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAQzB,MAAM,CAAC,MAAM,0BAA0B,GAAgC,CAAC,CAAC,MAAM,CAC7E;IACE,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,CAAC,0BAA0B,CAAC;IAChE,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,QAAQ,CACjD,4KAA4K,CAC7K;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,sHAAsH,CACvH;CACF,CACF,CAAC"}
@@ -7,6 +7,7 @@ export type BenchmarkStatusResponse = {
7
7
  name: string;
8
8
  status: TaskStatus;
9
9
  test_uuids?: Array<string> | null | undefined;
10
+ parallel_models?: boolean | null | undefined;
10
11
  evaluators?: Array<TestRunEvaluator> | null | undefined;
11
12
  model_results?: Array<ModelResult> | null | undefined;
12
13
  leaderboard_summary?: Array<{
@@ -1 +1 @@
1
- {"version":3,"file":"benchmarkstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B,MAAM,MAAM,uBAAuB,GAAG;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACrE,aAAa,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACpC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACzC,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CAoCvB,CAAC"}
1
+ {"version":3,"file":"benchmarkstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B,MAAM,MAAM,uBAAuB,GAAG;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,eAAe,CAAC,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACrE,aAAa,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACpC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACzC,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CAuCvB,CAAC"}
@@ -15,6 +15,7 @@ export const BenchmarkStatusResponse$zodSchema = z.object({
15
15
  .optional().describe("Leaderboard comparing the models, one row per model. Columns vary by benchmark: a `model` column plus pass/fail counts, latency, cost, and one score column per evaluator, keyed by evaluator name"),
16
16
  model_results: z.array(ModelResult$zodSchema).nullable().optional().describe("Results for each model"),
17
17
  name: z.string().describe("Name of the run. A run nobody has renamed shows its number instead, such as `Run 1` for a test run or `Benchmark 1` for a benchmark"),
18
+ parallel_models: z.boolean().nullable().optional().describe("How the models were run. `true` means several ran at a time, `false` means each one ran only after the one before it had finished"),
18
19
  share_token: z.string().nullable().optional().describe("Token for building the public share URL"),
19
20
  status: TaskStatus$zodSchema,
20
21
  stopped_early: z.boolean().default(false).describe("Whether any model's run stopped before starting every test case, after too many failed in a row"),
@@ -1 +1 @@
1
- {"version":3,"file":"benchmarkstatusresponse.js","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAiB/B,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC9C,kFAAkF,CACnF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,0BAA0B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAClE,QAAQ,CACP,4GAA4G,CAC7G;IACH,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;IACD,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;SACnE,QAAQ,EAAE,CAAC,QAAQ,CAClB,oMAAoM,CACrM;IACH,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC1E,wBAAwB,CACzB;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACvB,qIAAqI,CACtI;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,aAAa,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAChD,iGAAiG,CAClG;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,sBAAsB,CAAC;IACpD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,wDAAwD,CACzD;CACF,CAAC,CAAC"}
1
+ {"version":3,"file":"benchmarkstatusresponse.js","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAkB/B,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC9C,kFAAkF,CACnF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,0BAA0B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAClE,QAAQ,CACP,4GAA4G,CAC7G;IACH,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;IACD,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;SACnE,QAAQ,EAAE,CAAC,QAAQ,CAClB,oMAAoM,CACrM;IACH,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC1E,wBAAwB,CACzB;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACvB,qIAAqI,CACtI;IACD,eAAe,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACzD,mIAAmI,CACpI;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,aAAa,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAChD,iGAAiG,CAClG;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,sBAAsB,CAAC;IACpD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,wDAAwD,CACzD;CACF,CAAC,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@dalmia/calibrate-mcp",
3
- "version": "0.0.59",
3
+ "version": "0.0.60",
4
4
  "author": "Speakeasy",
5
5
  "type": "module",
6
6
  "sideEffects": false,
@@ -930,7 +930,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
930
930
  <h1>Instructions</h1>
931
931
  <p>One-click installation for Claude Desktop users</p>
932
932
  <div class="instruction-item">
933
- <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.59/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
933
+ <a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.60/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
934
934
  📥 Download MCP Bundle
935
935
  </a>
936
936
  </div>
package/src/lib/config.ts CHANGED
@@ -65,8 +65,8 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
65
65
  export const SDK_METADATA = {
66
66
  language: "typescript",
67
67
  openapiDocVersion: "0.1.0",
68
- sdkVersion: "0.0.59",
68
+ sdkVersion: "0.0.60",
69
69
  genVersion: "2.915.1",
70
70
  userAgent:
71
- "speakeasy-sdk/mcp-typescript 0.0.59 2.915.1 0.1.0 @dalmia/calibrate-mcp",
71
+ "speakeasy-sdk/mcp-typescript 0.0.60 2.915.1 0.1.0 @dalmia/calibrate-mcp",
72
72
  } as const;
@@ -21,7 +21,7 @@ const routes = buildRouteMap({
21
21
  export const app = buildApplication(routes, {
22
22
  name: "mcp",
23
23
  versionInfo: {
24
- currentVersion: "0.0.59",
24
+ currentVersion: "0.0.60",
25
25
  },
26
26
  });
27
27
 
@@ -73,7 +73,7 @@ export function createMCPServer(deps: {
73
73
  }) {
74
74
  const server = new McpServer({
75
75
  name: "CalibrateMcp",
76
- version: "0.0.59",
76
+ version: "0.0.60",
77
77
  });
78
78
 
79
79
  const getClient = deps.getSDK || (() =>
@@ -14,7 +14,7 @@ export const BenchmarkRequest$zodSchema: z.ZodType<BenchmarkRequest> = z.object(
14
14
  {
15
15
  models: z.array(z.string()).describe("Model names to benchmark"),
16
16
  parallel_models: z.boolean().default(true).describe(
17
- "Whether to run the models at the same time. Set false to run them one after another",
17
+ "How to run the models. `true` runs several at a time, `false` runs each one only after the one before it has finished. Use `false` to keep the load on your own agent down",
18
18
  ),
19
19
  test_uuids: z.array(z.string()).nullable().optional().describe(
20
20
  "A subset of the agent's linked tests to benchmark. Each ID must be linked to the agent. Omit to run all linked tests",
@@ -15,6 +15,7 @@ export type BenchmarkStatusResponse = {
15
15
  name: string;
16
16
  status: TaskStatus;
17
17
  test_uuids?: Array<string> | null | undefined;
18
+ parallel_models?: boolean | null | undefined;
18
19
  evaluators?: Array<TestRunEvaluator> | null | undefined;
19
20
  model_results?: Array<ModelResult> | null | undefined;
20
21
  leaderboard_summary?: Array<{ [k: string]: any }> | null | undefined;
@@ -51,6 +52,9 @@ export const BenchmarkStatusResponse$zodSchema: z.ZodType<
51
52
  name: z.string().describe(
52
53
  "Name of the run. A run nobody has renamed shows its number instead, such as `Run 1` for a test run or `Benchmark 1` for a benchmark",
53
54
  ),
55
+ parallel_models: z.boolean().nullable().optional().describe(
56
+ "How the models were run. `true` means several ran at a time, `false` means each one ran only after the one before it had finished",
57
+ ),
54
58
  share_token: z.string().nullable().optional().describe(
55
59
  "Token for building the public share URL",
56
60
  ),