@inspectr/mcplab 1.12.1 → 1.13.0
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.
- package/dist/app/assets/{index-Djwh_08m.js → index-BEjmzIAl.js} +68 -68
- package/dist/app/index.html +1 -1
- package/dist/app-server/result-assistant-domain.js +1 -1
- package/dist/app-server/result-assistant-domain.js.map +1 -1
- package/dist/app-server/result-assistant-tools.d.ts.map +1 -1
- package/dist/app-server/result-assistant-tools.js +3 -2
- package/dist/app-server/result-assistant-tools.js.map +1 -1
- package/dist/cli.js +159 -0
- package/dist/cli.js.map +1 -1
- package/dist/results/context.d.ts +3 -0
- package/dist/results/context.d.ts.map +1 -0
- package/dist/results/context.js +2 -0
- package/dist/results/context.js.map +1 -0
- package/dist/results/format.d.ts +21 -0
- package/dist/results/format.d.ts.map +1 -0
- package/dist/results/format.js +92 -0
- package/dist/results/format.js.map +1 -0
- package/dist/results/indexer.d.ts +9 -0
- package/dist/results/indexer.d.ts.map +1 -0
- package/dist/results/indexer.js +2 -0
- package/dist/results/indexer.js.map +1 -0
- package/dist/results/search.d.ts +2 -0
- package/dist/results/search.d.ts.map +1 -0
- package/dist/results/search.js +2 -0
- package/dist/results/search.js.map +1 -0
- package/dist/results/types.d.ts +11 -0
- package/dist/results/types.d.ts.map +1 -0
- package/dist/results/types.js +2 -0
- package/dist/results/types.js.map +1 -0
- package/dist/test-results-fixture.d.ts +7 -0
- package/dist/test-results-fixture.d.ts.map +1 -0
- package/dist/test-results-fixture.js +82 -0
- package/dist/test-results-fixture.js.map +1 -0
- package/package.json +6 -6
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"search.js","sourceRoot":"","sources":["../../src/results/search.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,QAAQ,EAAE,cAAc,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/results/types.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,YAAY,EACZ,YAAY,EACZ,SAAS,EACT,SAAS,EACT,aAAa,EACb,aAAa,EACb,cAAc,EACd,aAAa,EACd,MAAM,uBAAuB,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/results/types.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"test-results-fixture.d.ts","sourceRoot":"","sources":["../src/test-results-fixture.ts"],"names":[],"mappings":"AAKA,MAAM,MAAM,UAAU,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAE1E,wBAAgB,uBAAuB,IAAI,UAAU,CAwFpD"}
|
|
@@ -0,0 +1,82 @@
|
|
|
1
|
+
import { mkdtempSync, mkdirSync, writeFileSync } from 'node:fs';
|
|
2
|
+
import { tmpdir } from 'node:os';
|
|
3
|
+
import { join } from 'node:path';
|
|
4
|
+
export function createResultsRunFixture() {
|
|
5
|
+
const root = mkdtempSync(join(tmpdir(), 'mcplab-results-'));
|
|
6
|
+
const runsDir = join(root, 'mcplab', 'results', 'evaluation-runs');
|
|
7
|
+
const runId = '20260206-212239';
|
|
8
|
+
const runDir = join(runsDir, runId);
|
|
9
|
+
mkdirSync(runDir, { recursive: true });
|
|
10
|
+
const results = {
|
|
11
|
+
metadata: {
|
|
12
|
+
run_id: runId,
|
|
13
|
+
timestamp: '2026-02-06T21:22:39.000Z',
|
|
14
|
+
config_hash: 'abc123',
|
|
15
|
+
cli_version: '1.0.0',
|
|
16
|
+
mcp_server_versions: {}
|
|
17
|
+
},
|
|
18
|
+
summary: {
|
|
19
|
+
total_scenarios: 1,
|
|
20
|
+
total_runs: 1,
|
|
21
|
+
pass_rate: 0,
|
|
22
|
+
avg_tool_calls_per_run: 1,
|
|
23
|
+
avg_tool_latency_ms: 50
|
|
24
|
+
},
|
|
25
|
+
scenarios: [
|
|
26
|
+
{
|
|
27
|
+
scenario_id: 'search-tags',
|
|
28
|
+
agent: 'claude-haiku',
|
|
29
|
+
runs: [
|
|
30
|
+
{
|
|
31
|
+
run_index: 0,
|
|
32
|
+
pass: false,
|
|
33
|
+
failures: ['response assertion failed'],
|
|
34
|
+
tool_calls: ['search_tags'],
|
|
35
|
+
tool_call_count: 1,
|
|
36
|
+
tool_sequence: ['search_tags'],
|
|
37
|
+
tool_usage: { search_tags: 1 },
|
|
38
|
+
tool_durations_ms: [50],
|
|
39
|
+
final_text: 'Could not complete request because timeout',
|
|
40
|
+
extracted: {},
|
|
41
|
+
error: 'timeout'
|
|
42
|
+
}
|
|
43
|
+
],
|
|
44
|
+
pass_rate: 0,
|
|
45
|
+
distinct_sequences: { '["search_tags"]': 1 },
|
|
46
|
+
tool_usage_frequency: { search_tags: 1 },
|
|
47
|
+
extracted_values: {},
|
|
48
|
+
last_final_answer: 'Could not complete request because timeout'
|
|
49
|
+
}
|
|
50
|
+
]
|
|
51
|
+
};
|
|
52
|
+
writeFileSync(join(runDir, 'results.json'), `${JSON.stringify(results, null, 2)}\n`, 'utf8');
|
|
53
|
+
writeFileSync(join(runDir, 'summary.md'), '# MCP Eval Summary\n\n| Scenario | Agent | Runs | Pass rate |\n|---|---|---|---|\n| search-tags | claude-haiku | 1 | 0% |\n', 'utf8');
|
|
54
|
+
writeFileSync(join(runDir, 'trace.jsonl'), `${JSON.stringify({
|
|
55
|
+
type: 'scenario_run',
|
|
56
|
+
trace_version: 3,
|
|
57
|
+
run_index: 0,
|
|
58
|
+
scenario_id: 'search-tags',
|
|
59
|
+
agent: 'claude-haiku',
|
|
60
|
+
provider: 'anthropic',
|
|
61
|
+
model: 'claude-3-haiku',
|
|
62
|
+
ts_start: '2026-02-06T21:22:39.000Z',
|
|
63
|
+
ts_end: '2026-02-06T21:22:40.000Z',
|
|
64
|
+
pass: false,
|
|
65
|
+
messages: [
|
|
66
|
+
{
|
|
67
|
+
role: 'assistant',
|
|
68
|
+
content: [
|
|
69
|
+
{
|
|
70
|
+
type: 'tool_result',
|
|
71
|
+
tool_use_id: 'call_1',
|
|
72
|
+
name: 'search_tags',
|
|
73
|
+
content: [{ type: 'text', text: 'tool search_tags returned timeout after 5000ms' }],
|
|
74
|
+
is_error: true
|
|
75
|
+
}
|
|
76
|
+
]
|
|
77
|
+
}
|
|
78
|
+
]
|
|
79
|
+
})}\nnot-json\n`, 'utf8');
|
|
80
|
+
return { root, runsDir, runId };
|
|
81
|
+
}
|
|
82
|
+
//# sourceMappingURL=test-results-fixture.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"test-results-fixture.js","sourceRoot":"","sources":["../src/test-results-fixture.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAChE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAKjC,MAAM,UAAU,uBAAuB;IACrC,MAAM,IAAI,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,iBAAiB,CAAC,CAAC,CAAC;IAC5D,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,iBAAiB,CAAC,CAAC;IACnE,MAAM,KAAK,GAAG,iBAAiB,CAAC;IAChC,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;IACpC,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAEvC,MAAM,OAAO,GAAgB;QAC3B,QAAQ,EAAE;YACR,MAAM,EAAE,KAAK;YACb,SAAS,EAAE,0BAA0B;YACrC,WAAW,EAAE,QAAQ;YACrB,WAAW,EAAE,OAAO;YACpB,mBAAmB,EAAE,EAAE;SACxB;QACD,OAAO,EAAE;YACP,eAAe,EAAE,CAAC;YAClB,UAAU,EAAE,CAAC;YACb,SAAS,EAAE,CAAC;YACZ,sBAAsB,EAAE,CAAC;YACzB,mBAAmB,EAAE,EAAE;SACxB;QACD,SAAS,EAAE;YACT;gBACE,WAAW,EAAE,aAAa;gBAC1B,KAAK,EAAE,cAAc;gBACrB,IAAI,EAAE;oBACJ;wBACE,SAAS,EAAE,CAAC;wBACZ,IAAI,EAAE,KAAK;wBACX,QAAQ,EAAE,CAAC,2BAA2B,CAAC;wBACvC,UAAU,EAAE,CAAC,aAAa,CAAC;wBAC3B,eAAe,EAAE,CAAC;wBAClB,aAAa,EAAE,CAAC,aAAa,CAAC;wBAC9B,UAAU,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE;wBAC9B,iBAAiB,EAAE,CAAC,EAAE,CAAC;wBACvB,UAAU,EAAE,4CAA4C;wBACxD,SAAS,EAAE,EAAE;wBACb,KAAK,EAAE,SAAS;qBACjB;iBACF;gBACD,SAAS,EAAE,CAAC;gBACZ,kBAAkB,EAAE,EAAE,iBAAiB,EAAE,CAAC,EAAE;gBAC5C,oBAAoB,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE;gBACxC,gBAAgB,EAAE,EAAE;gBACpB,iBAAiB,EAAE,4CAA4C;aAChE;SACF;KACF,CAAC;IAEF,aAAa,CAAC,IAAI,CAAC,MAAM,EAAE,cAAc,CAAC,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC7F,aAAa,CACX,IAAI,CAAC,MAAM,EAAE,YAAY,CAAC,EAC1B,6HAA6H,EAC7H,MAAM,CACP,CAAC;IACF,aAAa,CACX,IAAI,CAAC,MAAM,EAAE,aAAa,CAAC,EAC3B,GAAG,IAAI,CAAC,SAAS,CAAC;QAChB,IAAI,EAAE,cAAc;QACpB,aAAa,EAAE,CAAC;QAChB,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,aAAa;QAC1B,KAAK,EAAE,cAAc;QACrB,QAAQ,EAAE,WAAW;QACrB,KAAK,EAAE,gBAAgB;QACvB,QAAQ,EAAE,0BAA0B;QACpC,MAAM,EAAE,0BAA0B;QAClC,IAAI,EAAE,KAAK;QACX,QAAQ,EAAE;YACR;gBACE,IAAI,EAAE,WAAW;gBACjB,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,aAAa;wBACnB,WAAW,EAAE,QAAQ;wBACrB,IAAI,EAAE,aAAa;wBACnB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,gDAAgD,EAAE,CAAC;wBACnF,QAAQ,EAAE,IAAI;qBACf;iBACF;aACF;SACF;KACF,CAAC,cAAc,EAChB,MAAM,CACP,CAAC;IAEF,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AAClC,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@inspectr/mcplab",
|
|
3
|
-
"version": "1.
|
|
3
|
+
"version": "1.13.0",
|
|
4
4
|
"description": "MCPLab - Test and evaluate MCP servers with LLMs — run evals, compare agents and launch the MCPLab web app",
|
|
5
5
|
"license": "MIT",
|
|
6
6
|
"homepage": "https://mcplab.inspectr.dev",
|
|
@@ -44,12 +44,12 @@
|
|
|
44
44
|
"vitest": "^3.2.4"
|
|
45
45
|
},
|
|
46
46
|
"dependencies": {
|
|
47
|
-
"@inspectr/mcplab-core": "1.
|
|
48
|
-
"@inspectr/mcplab-mcp-server": "1.
|
|
49
|
-
"@inspectr/mcplab-reporting": "1.1.
|
|
47
|
+
"@inspectr/mcplab-core": "1.10.0",
|
|
48
|
+
"@inspectr/mcplab-mcp-server": "1.3.0",
|
|
49
|
+
"@inspectr/mcplab-reporting": "1.1.10",
|
|
50
50
|
"commander": "^12.1.0",
|
|
51
|
-
"dotenv": "^16.
|
|
51
|
+
"dotenv": "^16.6.1",
|
|
52
52
|
"kleur": "^4.1.5",
|
|
53
|
-
"yaml": "^2.
|
|
53
|
+
"yaml": "^2.8.4"
|
|
54
54
|
}
|
|
55
55
|
}
|