File size: 25,575 Bytes
c6856c9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
#!/usr/bin/env python3
from __future__ import annotations

import argparse
import ast
import csv
import importlib.util
import inspect
import json
import multiprocessing as mp
import os
import random
import re
import shutil
import subprocess
import sys
import tempfile
import traceback
from concurrent.futures import ThreadPoolExecutor, as_completed
from collections import Counter, defaultdict
from pathlib import Path
from typing import Any


PROJECT_ROOT = Path(__file__).resolve().parents[2]
DEFAULT_MCP_ROOT = PROJECT_ROOT / "biomni_web" / "backend" / "data" / "mcp_generated"
DEFAULT_CATALOG = PROJECT_ROOT / "graph_outputs" / "mcp_generated_graph_all_20260522_124110" / "server_catalog.json"
DEFAULT_OUTPUT_DIR = PROJECT_ROOT / "experiments" / "execution_audit" / "results"


def normalize_server_name(server_dir: Path) -> str:
    name = server_dir.name
    return name[4:] if name.startswith("mcp_") else name


def safe_name(name: str) -> str:
    return re.sub(r"[^0-9a-zA-Z_]", "_", name)


def classify_error(stderr: str, stdout: str = "", returncode: int | None = None) -> str:
    text = f"{stderr}\n{stdout}".lower()
    if returncode == 124 or "timeout" in text or "timed out" in text:
        return "timeout"
    if "no module named" in text or "modulenotfounderror" in text or "importerror" in text:
        return "missing_python_dependency"
    if "no such file or directory" in text or "not found" in text or "filenotfounderror" in text:
        return "missing_executable_or_file"
    if "syntaxerror" in text or "indentationerror" in text:
        return "syntax_error"
    if "validation error" in text or "schema" in text or "missing required" in text:
        return "schema_or_argument_error"
    if "calledprocesserror" in text or "non-zero exit" in text or "exit status" in text:
        return "runtime_nonzero_exit"
    if "traceback" in text or "exception" in text or "error" in text:
        return "runtime_exception"
    if returncode not in (None, 0):
        return "nonzero_exit"
    return "none"


def run_cmd(cmd: list[str], *, timeout: float, cwd: Path | None = None) -> dict[str, Any]:
    try:
        proc = subprocess.run(
            cmd,
            cwd=str(cwd) if cwd else None,
            text=True,
            capture_output=True,
            timeout=timeout,
        )
        return {
            "returncode": proc.returncode,
            "stdout": proc.stdout[-4000:],
            "stderr": proc.stderr[-4000:],
            "timed_out": False,
        }
    except subprocess.TimeoutExpired as exc:
        return {
            "returncode": 124,
            "stdout": (exc.stdout or "")[-4000:] if isinstance(exc.stdout, str) else "",
            "stderr": (exc.stderr or "")[-4000:] if isinstance(exc.stderr, str) else "timeout",
            "timed_out": True,
        }


def server_files(server_dir: Path) -> tuple[Path | None, Path | None]:
    app_dir = server_dir / "app"
    shim = next(iter(sorted(app_dir.glob("*_shim_server.py"))), None)
    raw = next(
        iter(sorted(p for p in app_dir.glob("*_server.py") if not p.name.endswith("_shim_server.py"))),
        None,
    )
    return shim, raw


def audit_syntax(path: Path, python: str, timeout: float) -> dict[str, Any]:
    result = run_cmd([python, "-m", "py_compile", str(path)], timeout=timeout)
    ok = result["returncode"] == 0
    return {
        "ok": ok,
        "error_type": "none" if ok else classify_error(result["stderr"], result["stdout"], result["returncode"]),
        "stderr": result["stderr"],
    }


def _import_worker(path: str, queue: mp.Queue) -> None:
    try:
        module_name = f"_mcp_audit_{safe_name(Path(path).stem)}_{os.getpid()}"
        spec = importlib.util.spec_from_file_location(module_name, path)
        if spec is None or spec.loader is None:
            raise RuntimeError("cannot build import spec")
        module = importlib.util.module_from_spec(spec)
        spec.loader.exec_module(module)
        queue.put({"ok": True, "error": ""})
    except BaseException:
        queue.put({"ok": False, "error": traceback.format_exc()[-4000:]})


def run_import(path: Path, timeout: float) -> dict[str, Any]:
    ctx = mp.get_context("spawn")
    queue: mp.Queue = ctx.Queue()
    proc = ctx.Process(target=_import_worker, args=(str(path), queue))
    proc.start()
    proc.join(timeout)
    if proc.is_alive():
        proc.terminate()
        proc.join(1)
        return {"ok": False, "error_type": "timeout", "stderr": "import timed out"}
    payload = queue.get() if not queue.empty() else {"ok": proc.exitcode == 0, "error": ""}
    ok = bool(payload.get("ok"))
    err = payload.get("error", "")
    return {
        "ok": ok,
        "error_type": "none" if ok else classify_error(err, returncode=proc.exitcode),
        "stderr": err,
    }


def audit_import_subprocess(path: Path, python: str, timeout: float) -> dict[str, Any]:
    code = (
        "import importlib.util, pathlib; "
        f"p=pathlib.Path({str(path)!r}); "
        "spec=importlib.util.spec_from_file_location('_mcp_audit_mod', str(p)); "
        "m=importlib.util.module_from_spec(spec); "
        "spec.loader.exec_module(m)"
    )
    result = run_cmd([python, "-c", code], timeout=timeout)
    ok = result["returncode"] == 0
    return {
        "ok": ok,
        "error_type": "none" if ok else classify_error(result["stderr"], result["stdout"], result["returncode"]),
        "stderr": result["stderr"],
    }


def audit_startup(path: Path, python: str, timeout: float, grace: float) -> dict[str, Any]:
    proc = subprocess.Popen(
        [python, str(path)],
        stdin=subprocess.PIPE,
        stdout=subprocess.PIPE,
        stderr=subprocess.PIPE,
        text=True,
    )
    try:
        stdout, stderr = proc.communicate(timeout=grace)
        ok = proc.returncode == 0
        return {
            "ok": ok,
            "returncode": proc.returncode,
            "error_type": "none" if ok else classify_error(stderr, stdout, proc.returncode),
            "stderr": (stderr or "")[-4000:],
        }
    except subprocess.TimeoutExpired:
        proc.terminate()
        try:
            stdout, stderr = proc.communicate(timeout=timeout)
        except subprocess.TimeoutExpired:
            proc.kill()
            stdout, stderr = proc.communicate(timeout=1)
        # A stdio MCP server should keep running while it waits for requests.
        if not (stderr or "").strip():
            return {"ok": True, "returncode": 124, "error_type": "none", "stderr": ""}
        return {
            "ok": False,
            "returncode": 124,
            "error_type": classify_error(stderr, stdout, 124),
            "stderr": (stderr or "")[-4000:],
        }


def extract_command_literals(source_path: Path) -> list[str]:
    try:
        text = source_path.read_text(encoding="utf-8")
    except OSError:
        return []
    candidates: list[str] = []
    for match in re.finditer(r"(?:command|cmd)\s*(?::[^=]+)?=\s*\[\s*['\"]([^'\"]+)['\"]", text):
        candidates.append(match.group(1))
    for match in re.finditer(r"subprocess\.(?:run|Popen|check_output|check_call)\(\s*\[\s*['\"]([^'\"]+)['\"]", text):
        candidates.append(match.group(1))
    cleaned = []
    for item in candidates:
        base = Path(item).name
        if base and base not in {"python", "python3", "Rscript", "bash", "sh"}:
            cleaned.append(base)
    return sorted(set(cleaned))


def audit_help_version(commands: list[str], timeout: float) -> dict[str, Any]:
    if not commands:
        return {"probed": False, "ok": False, "command": "", "flag": "", "error_type": "no_command_literal"}
    for command in commands:
        if shutil.which(command) is None:
            continue
        for flag in ("--help", "-h", "--version", "-version", "-v"):
            result = run_cmd([command, flag], timeout=timeout)
            combined = f"{result['stdout']}\n{result['stderr']}".strip()
            if result["returncode"] == 0 or combined:
                return {
                    "probed": True,
                    "ok": True,
                    "command": command,
                    "flag": flag,
                    "returncode": result["returncode"],
                    "error_type": "none",
                }
    first = commands[0]
    return {
        "probed": True,
        "ok": False,
        "command": first,
        "flag": "",
        "error_type": "missing_executable_or_file",
    }


def load_catalog(path: Path) -> list[dict[str, Any]]:
    return json.loads(path.read_text(encoding="utf-8"))


def stratified_tool_sample(catalog: list[dict[str, Any]], sample_size: int, seed: int) -> list[dict[str, Any]]:
    by_category: dict[str, list[dict[str, Any]]] = defaultdict(list)
    for server in catalog:
        for tool in server.get("tools", []):
            by_category[server.get("category", "unknown")].append(
                {
                    "server": server["name"],
                    "category": server.get("category", "unknown"),
                    "tool": tool.get("name"),
                    "schema": tool.get("inputSchema", {}),
                }
            )
    rng = random.Random(seed)
    total = sum(len(items) for items in by_category.values())
    selected: list[dict[str, Any]] = []
    for category, items in sorted(by_category.items()):
        quota = max(1, round(sample_size * len(items) / total))
        rng.shuffle(items)
        selected.extend(items[:quota])
    if len(selected) > sample_size:
        selected = selected[:sample_size]
    elif len(selected) < sample_size:
        seen = {(x["server"], x["tool"]) for x in selected}
        rest = [
            item
            for items in by_category.values()
            for item in items
            if (item["server"], item["tool"]) not in seen
        ]
        rng.shuffle(rest)
        selected.extend(rest[: sample_size - len(selected)])
    return selected


def make_fixture_files(tmp: Path) -> dict[str, Path]:
    files = {
        "fasta": tmp / "tiny.fa",
        "fastq": tmp / "tiny.fastq",
        "csv": tmp / "tiny.csv",
        "tsv": tmp / "tiny.tsv",
        "txt": tmp / "tiny.txt",
        "vcf": tmp / "tiny.vcf",
        "bam": tmp / "tiny.bam",
        "gff": tmp / "tiny.gff",
        "json": tmp / "tiny.json",
    }
    files["fasta"].write_text(">seq1\nACGTACGTACGTACGTACGT\n", encoding="utf-8")
    files["fastq"].write_text("@seq1\nACGTACGTACGT\n+\nFFFFFFFFFFFF\n", encoding="utf-8")
    files["csv"].write_text("gene,count\nA,1\nB,2\n", encoding="utf-8")
    files["tsv"].write_text("gene\tcount\nA\t1\nB\t2\n", encoding="utf-8")
    files["txt"].write_text("tiny fixture\n", encoding="utf-8")
    files["vcf"].write_text("##fileformat=VCFv4.2\n#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO\n1\t1\t.\tA\tC\t.\t.\t.\n", encoding="utf-8")
    files["bam"].write_bytes(b"BAM\1")
    files["gff"].write_text("##gff-version 3\nseq1\t.\tgene\t1\t10\t.\t+\t.\tID=g1\n", encoding="utf-8")
    files["json"].write_text('{"items": [1, 2, 3]}\n', encoding="utf-8")
    return files


def value_for_arg(name: str, schema: dict[str, Any], fixtures: dict[str, Path], tmp: Path) -> Any:
    lname = name.lower()
    typ = (schema or {}).get("type", "string")
    if any(k in lname for k in ("out", "output", "report", "result", "prefix", "log")):
        return str(tmp / f"{safe_name(name)}.out")
    if "fastq" in lname or "fq" in lname:
        return str(fixtures["fastq"])
    if "fasta" in lname or "fa" == lname or lname.endswith("_fa") or "reference" in lname or "query" in lname:
        return str(fixtures["fasta"])
    if "vcf" in lname:
        return str(fixtures["vcf"])
    if "bam" in lname or "sam" in lname:
        return str(fixtures["bam"])
    if "gff" in lname or "gtf" in lname or "annotation" in lname:
        return str(fixtures["gff"])
    if "json" in lname:
        return str(fixtures["json"])
    if "csv" in lname or "table" in lname or "matrix" in lname or "count" in lname:
        return str(fixtures["csv"])
    if "tsv" in lname:
        return str(fixtures["tsv"])
    if "file" in lname or "path" in lname or "input" in lname or "contig" in lname or "reads" in lname:
        return str(fixtures["fasta"])
    if typ == "boolean":
        return False
    if typ == "number" or typ == "integer":
        return 1
    if typ == "array":
        return [str(fixtures["fasta"])]
    if typ == "object":
        return {}
    return "tiny"


def find_source_for_server(mcp_root: Path, server: str) -> Path | None:
    server_dir = mcp_root / f"mcp_{server}"
    shim, raw = server_files(server_dir)
    return shim or raw


def _tool_worker(source: str, tool_name: str, kwargs: dict[str, Any], queue: mp.Queue) -> None:
    try:
        path = Path(source)
        module_name = f"_mcp_tool_audit_{safe_name(path.stem)}_{os.getpid()}"
        spec = importlib.util.spec_from_file_location(module_name, str(path))
        if spec is None or spec.loader is None:
            raise RuntimeError("cannot build import spec")
        module = importlib.util.module_from_spec(spec)
        spec.loader.exec_module(module)
        fn = getattr(module, tool_name, None)
        if fn is None and hasattr(module, "_load_functions"):
            for candidate in module._load_functions():
                if getattr(candidate, "__name__", "") == tool_name:
                    fn = candidate
                    break
        if fn is None:
            raise AttributeError(f"tool function not found: {tool_name}")
        signature = inspect.signature(fn)
        coerced_kwargs = {}
        for key, value in kwargs.items():
            annotation = signature.parameters.get(key).annotation if key in signature.parameters else inspect._empty
            annotation_text = str(annotation)
            if value is not None and ("Path" in annotation_text or annotation is Path):
                coerced_kwargs[key] = Path(value)
            else:
                coerced_kwargs[key] = value
        kwargs = coerced_kwargs
        result = fn(**kwargs)
        queue.put({"ok": True, "result_type": type(result).__name__, "error": ""})
    except BaseException:
        queue.put({"ok": False, "result_type": "", "error": traceback.format_exc()[-4000:]})


def audit_tool_call(mcp_root: Path, item: dict[str, Any], timeout: float) -> dict[str, Any]:
    source = find_source_for_server(mcp_root, item["server"])
    if source is None:
        return {**item, "ok": False, "error_type": "missing_executable_or_file", "stderr": "server source not found"}
    schema = item.get("schema") or {}
    properties = schema.get("properties") or {}
    required = schema.get("required") or []
    with tempfile.TemporaryDirectory(prefix="mcp_tool_audit_") as tmp_s:
        tmp = Path(tmp_s)
        fixtures = make_fixture_files(tmp)
        kwargs = {name: value_for_arg(name, properties.get(name, {}), fixtures, tmp) for name in required}
        ctx = mp.get_context("spawn")
        queue: mp.Queue = ctx.Queue()
        proc = ctx.Process(target=_tool_worker, args=(str(source), item["tool"], kwargs, queue))
        proc.start()
        proc.join(timeout)
        if proc.is_alive():
            proc.terminate()
            proc.join(1)
            return {**item, "ok": False, "error_type": "timeout", "stderr": "tool call timed out", "kwargs": kwargs}
        payload = queue.get() if not queue.empty() else {"ok": proc.exitcode == 0, "error": ""}
        ok = bool(payload.get("ok"))
        err = payload.get("error", "")
        return {
            **item,
            "ok": ok,
            "error_type": "none" if ok else classify_error(err, returncode=proc.exitcode),
            "stderr": err,
            "kwargs": kwargs,
            "result_type": payload.get("result_type", ""),
        }


def audit_one_server(payload: tuple[str, str, float, float, float, float, float]) -> dict[str, Any]:
    server_dir_s, python, syntax_timeout, import_timeout, startup_timeout, startup_grace, help_timeout = payload
    server_dir = Path(server_dir_s)
    server = normalize_server_name(server_dir)
    shim, raw = server_files(server_dir)
    target = shim or raw
    if target is None:
        return {"server": server, "has_source": False}
    syntax = audit_syntax(target, python, syntax_timeout)
    imported = (
        audit_import_subprocess(target, python, import_timeout)
        if syntax["ok"]
        else {"ok": False, "error_type": syntax["error_type"], "stderr": syntax["stderr"]}
    )
    startup = (
        audit_startup(target, python, startup_timeout, startup_grace)
        if syntax["ok"]
        else {"ok": False, "error_type": syntax["error_type"], "stderr": syntax["stderr"]}
    )
    commands = extract_command_literals(raw or target)
    help_version = audit_help_version(commands, help_timeout)
    return {
        "server": server,
        "source": str(target),
        "has_source": True,
        "syntax_ok": syntax["ok"],
        "import_ok": imported["ok"],
        "startup_ok": startup["ok"],
        "help_version_probed": help_version["probed"],
        "help_version_ok": help_version["ok"],
        "help_version_command": help_version.get("command", ""),
        "help_version_flag": help_version.get("flag", ""),
        "syntax_error_type": syntax["error_type"],
        "import_error_type": imported["error_type"],
        "startup_error_type": startup["error_type"],
        "help_version_error_type": help_version["error_type"],
    }


def pct(n: int, d: int) -> str:
    return f"{(100 * n / d):.1f}%" if d else "n/a"


def write_csv(path: Path, rows: list[dict[str, Any]]) -> None:
    if not rows:
        path.write_text("", encoding="utf-8")
        return
    fieldnames = sorted({key for row in rows for key in row})
    with path.open("w", newline="", encoding="utf-8") as handle:
        writer = csv.DictWriter(handle, fieldnames=fieldnames)
        writer.writeheader()
        writer.writerows(rows)


def main() -> None:
    parser = argparse.ArgumentParser(description="Execution audit for generated BioinfoMCP servers.")
    parser.add_argument("--mcp-root", type=Path, default=DEFAULT_MCP_ROOT)
    parser.add_argument("--catalog", type=Path, default=DEFAULT_CATALOG)
    parser.add_argument("--output-dir", type=Path, default=DEFAULT_OUTPUT_DIR)
    parser.add_argument("--python", default=sys.executable)
    parser.add_argument("--sample-tools", type=int, default=300)
    parser.add_argument("--seed", type=int, default=13)
    parser.add_argument("--syntax-timeout", type=float, default=8)
    parser.add_argument("--import-timeout", type=float, default=8)
    parser.add_argument("--startup-grace", type=float, default=2)
    parser.add_argument("--startup-timeout", type=float, default=1)
    parser.add_argument("--help-timeout", type=float, default=4)
    parser.add_argument("--tool-timeout", type=float, default=8)
    parser.add_argument("--server-workers", type=int, default=16)
    parser.add_argument("--tool-workers", type=int, default=8)
    args = parser.parse_args()

    args.output_dir.mkdir(parents=True, exist_ok=True)
    server_dirs = sorted(p for p in args.mcp_root.glob("mcp_*") if p.is_dir())
    server_rows: list[dict[str, Any]] = []

    server_payloads = [
        (
            str(server_dir),
            args.python,
            args.syntax_timeout,
            args.import_timeout,
            args.startup_timeout,
            args.startup_grace,
            args.help_timeout,
        )
        for server_dir in server_dirs
    ]
    with mp.Pool(processes=max(1, args.server_workers)) as pool:
        for index, row in enumerate(pool.imap_unordered(audit_one_server, server_payloads), start=1):
            server_rows.append(row)
            if index % 100 == 0:
                print(f"audited {index}/{len(server_dirs)} servers", flush=True)
    server_rows.sort(key=lambda row: row.get("server", ""))

    catalog = load_catalog(args.catalog)
    sample = stratified_tool_sample(catalog, args.sample_tools, args.seed)
    tool_rows: list[dict[str, Any]] = []
    with ThreadPoolExecutor(max_workers=max(1, args.tool_workers)) as executor:
        futures = [executor.submit(audit_tool_call, args.mcp_root, item, args.tool_timeout) for item in sample]
        for index, future in enumerate(as_completed(futures), start=1):
            tool_rows.append(future.result())
            if index % 25 == 0:
                print(f"audited {index}/{len(sample)} sampled tools", flush=True)
    tool_rows.sort(key=lambda row: (row.get("category", ""), row.get("server", ""), row.get("tool", "")))

    server_total = len(server_rows)
    source_present_total = sum(1 for r in server_rows if r.get("has_source"))
    tool_total_catalog = sum(len(s.get("tools", [])) for s in catalog)
    summary_rows = [
        {
            "audit_layer": "Generated server entrypoint present",
            "unit": "mcp_generated server directories",
            "n_tested": server_total,
            "n_success": source_present_total,
            "success_rate": pct(source_present_total, server_total),
            "notes": "server directory contains *_shim_server.py or *_server.py under app/",
        },
        {
            "audit_layer": "Server syntax/compile",
            "unit": "MCP servers",
            "n_tested": source_present_total,
            "n_success": sum(1 for r in server_rows if r.get("syntax_ok")),
            "success_rate": pct(sum(1 for r in server_rows if r.get("syntax_ok")), source_present_total),
            "notes": "python -m py_compile on generated shim/raw server",
        },
        {
            "audit_layer": "Server import",
            "unit": "MCP servers",
            "n_tested": sum(1 for r in server_rows if r.get("syntax_ok")),
            "n_success": sum(1 for r in server_rows if r.get("import_ok")),
            "success_rate": pct(
                sum(1 for r in server_rows if r.get("import_ok")),
                sum(1 for r in server_rows if r.get("syntax_ok")),
            ),
            "notes": "import generated server module in isolated process",
        },
        {
            "audit_layer": "MCP startup",
            "unit": "MCP servers",
            "n_tested": sum(1 for r in server_rows if r.get("syntax_ok")),
            "n_success": sum(1 for r in server_rows if r.get("startup_ok")),
            "success_rate": pct(
                sum(1 for r in server_rows if r.get("startup_ok")),
                sum(1 for r in server_rows if r.get("syntax_ok")),
            ),
            "notes": "stdio server starts and does not immediately traceback; waiting servers count as success",
        },
        {
            "audit_layer": "Help/version probe",
            "unit": "servers with inferred underlying CLI",
            "n_tested": sum(1 for r in server_rows if r.get("help_version_probed")),
            "n_success": sum(1 for r in server_rows if r.get("help_version_ok")),
            "success_rate": pct(
                sum(1 for r in server_rows if r.get("help_version_ok")),
                sum(1 for r in server_rows if r.get("help_version_probed")),
            ),
            "notes": "run inferred command with --help/-h/--version when command literal is found",
        },
        {
            "audit_layer": "Tiny fixture execution",
            "unit": "stratified sampled tools",
            "n_tested": len(tool_rows),
            "n_success": sum(1 for r in tool_rows if r.get("ok")),
            "success_rate": pct(sum(1 for r in tool_rows if r.get("ok")), len(tool_rows)),
            "notes": f"sampled from {tool_total_catalog} catalog tools; required args filled with tiny FASTA/FASTQ/CSV/VCF fixtures",
        },
    ]

    server_error_counts = Counter()
    for row in server_rows:
        if not row.get("has_source"):
            server_error_counts["missing_server_entrypoint"] += 1
        for key in ("syntax_error_type", "import_error_type", "startup_error_type", "help_version_error_type"):
            value = row.get(key)
            if value and value != "none" and value != "no_command_literal":
                server_error_counts[value] += 1
    tool_error_counts = Counter(r.get("error_type", "none") for r in tool_rows if not r.get("ok"))

    error_rows = []
    for name, count in server_error_counts.most_common():
        error_rows.append({"scope": "server", "error_type": name, "count": count})
    for name, count in tool_error_counts.most_common():
        error_rows.append({"scope": "sampled_tool", "error_type": name, "count": count})

    write_csv(args.output_dir / "server_audit.csv", server_rows)
    write_csv(args.output_dir / "tool_fixture_audit.csv", tool_rows)
    write_csv(args.output_dir / "summary_table.csv", summary_rows)
    write_csv(args.output_dir / "error_taxonomy.csv", error_rows)
    (args.output_dir / "summary.json").write_text(
        json.dumps(
            {
                "mcp_root": str(args.mcp_root),
                "catalog": str(args.catalog),
                "server_count": server_total,
                "catalog_tool_count": tool_total_catalog,
                "sampled_tool_count": len(tool_rows),
                "summary_table": summary_rows,
                "error_taxonomy": error_rows,
            },
            ensure_ascii=False,
            indent=2,
        ),
        encoding="utf-8",
    )
    print(json.dumps(summary_rows, indent=2), flush=True)
    print(f"wrote audit results to {args.output_dir}", flush=True)


if __name__ == "__main__":
    main()