MCP Observatory
4m ago Tracked 60,021 Last 4m ago 24h 2,474
github Python analyzed 2cc9449

klei30/tuner-mcp

github

Agent-native MCP control plane for Tinker post-training workflows: datasets, 34 Cookbook recipes, training, evaluation, checkpoints, and exports.

maintainer
klei30
licence
Apache-2.0
first seen
2026-09-13
last seen
2026-09-17
releases · 30d
2
short id

Drift inferred · capture-to-capture

No drift recorded — single capability capture; advisories appear once its surface changes.

capabilities 48 tools
transport stdio · streamable-http counts 48 tools · 0 res · 0 prompts permission surface via code analysis

tools

  • benchmarks_list
  • capabilities_get
  • checkpoint_delete
  • checkpoint_export
  • checkpoint_get
  • checkpoint_list
  • checkpoint_publish
  • checkpoint_set_ttl
  • checkpoint_unpublish
  • compare_runs
  • compute_logprobs
  • dataset_fetch_hf
  • dataset_inspect
  • dataset_prepare
  • dataset_probe_hf
  • dataset_render_preview
  • dataset_search_hf
  • dataset_validate
  • evaluate
  • evaluation_failures
  • evaluation_get
  • experiment_artifacts
  • experiment_autoplan
  • experiment_rollouts
  • models_list
  • object_get
  • objects_list
  • recipe_get
  • recipe_plan
  • recipe_start
  • recipes_list
  • sample
  • session_get
  • session_trace_export
  • sessions_list
  • train_distill
  • train_dpo
  • train_rl
  • train_sft
  • training_get
  • training_list
  • training_logs
  • training_metrics
  • training_plan
  • training_resume
  • training_start
  • training_stop
  • usage_get
skills & danger signals github-tarball
prompt-surface shipped agent-instruction files + hidden-content / dangerous-code findings — quoted from the analyzed source

analyzed commit 2cc9449 · analyzer v33 · 2d ago

skills & prompt files 1

danger signals1

code evidence vv0.1.1 · github-tarball
evidence-backed findings quoted directly from the published source artifact — not inferred

code files: 51

filesystem 27

  • fs (weak) klei30-tuner-mcp-2cc9449/scripts/smoke_docker_mcp.py :7 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/scripts/update_tool_schema.py :5 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/adapters.py :8 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/artifacts.py :4 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/control.py :8 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/datasets.py :14 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/discovery.py :8 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/evaluation.py :7 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :5 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/operations.py :5 import shutil
  • fs klei30-tuner-mcp-2cc9449/src/tuner/recipe_policy.py :3 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/settings.py :5 from pathlib import Path
show 15 more
  • fs klei30-tuner-mcp-2cc9449/src/tuner/store.py :10 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/worker_job.py :13 from pathlib import Path
  • fs klei30-tuner-mcp-2cc9449/src/tuner/workflows.py :7 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_artifacts.py :23 with path.open("ab") as handle:
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_datasets.py :2 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_experience_regressions.py :4 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_exports.py :8 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_hf_fetch.py :8 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_not_found.py :5 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_planning.py :2 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_recipe_catalog.py :2 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_safety_regressions.py :4 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_server.py :2 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_skill.py :1 from pathlib import Path
  • fs (weak) klei30-tuner-mcp-2cc9449/tests/test_store.py :4 from pathlib import Path

shell / exec 2

  • shell klei30-tuner-mcp-2cc9449/src/tuner/discovery.py :6 import subprocess
  • shell klei30-tuner-mcp-2cc9449/src/tuner/worker_job.py :9 import subprocess

network 3

  • net klei30-tuner-mcp-2cc9449/src/tuner/datasets.py :8 import urllib.parse
  • net klei30-tuner-mcp-2cc9449/src/tuner/operations.py :8 import urllib.request
  • net (weak) klei30-tuner-mcp-2cc9449/tests/test_hf_fetch.py :600 monkeypatch.setattr("urllib.request.urlopen", boom)

secrets 3

  • secrets (weak) klei30-tuner-mcp-2cc9449/scripts/check_http.py :14 token = os.environ.get("TUNER_MCP_AUTH_TOKEN")
  • secrets (weak) klei30-tuner-mcp-2cc9449/scripts/smoke_docker_mcp.py :14 api_key = os.environ.get("TINKER_API_KEY")
  • secrets klei30-tuner-mcp-2cc9449/src/tuner/settings.py :30 auth_token: str | None = field(default_factory=lambda: os.getenv("TUNER_AUTH_TOKEN") or None)

database 1

  • db klei30-tuner-mcp-2cc9449/src/tuner/store.py :6 import sqlite3

tool registrations 48

  • recipes_list klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :163
  • recipe_get klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :168
  • dataset_prepare klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :176
  • dataset_fetch_hf klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :184
  • dataset_search_hf klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :192
  • dataset_probe_hf klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :206
  • experiment_autoplan klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :214
  • training_plan klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :236
  • recipe_plan klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :244
  • training_start klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :259
  • recipe_start klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :272
  • training_stop klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :285
show 28 more
  • capabilities_get klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :293
  • models_list klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :343
  • dataset_validate klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :370
  • dataset_inspect klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :378
  • training_list klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :388
  • training_get klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :405
  • training_metrics klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :424
  • training_logs klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :459
  • checkpoint_list klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :501
  • checkpoint_get klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :510
  • sample klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :523
  • compute_logprobs klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :551
  • train_sft klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :578
  • evaluate klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :596
  • compare_runs klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :614
  • training_resume klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :641
  • train_dpo klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :674
  • train_rl klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :684
  • train_distill klei30-tuner-mcp-2cc9449/src/tuner/mcp_server.py :694
  • objects_list klei30-tuner-mcp-2cc9449/src/tuner/operations.py :335
  • object_get klei30-tuner-mcp-2cc9449/src/tuner/operations.py :362
  • dataset_render_preview klei30-tuner-mcp-2cc9449/src/tuner/operations.py :375
  • experiment_artifacts klei30-tuner-mcp-2cc9449/src/tuner/operations.py :412
  • experiment_rollouts klei30-tuner-mcp-2cc9449/src/tuner/operations.py :429
  • benchmarks_list klei30-tuner-mcp-2cc9449/src/tuner/operations.py :444
  • evaluation_get klei30-tuner-mcp-2cc9449/src/tuner/operations.py :461
  • evaluation_failures klei30-tuner-mcp-2cc9449/src/tuner/operations.py :469
  • usage_get klei30-tuner-mcp-2cc9449/src/tuner/operations.py :484

8 more not shown — this panel samples each group; the count above is the real total.

declared dependencies 7

  • fastmcp@==4.0.3
  • datasets@>=4.5,<5
  • ifbench@*
  • pydantic@>=2.11,<3
  • tinker@>=0.27.2,<0.28
  • tinker-cookbook@*
  • torch@>=2.10