local-coding-agent 1.1.0__tar.gz → 1.1.1__tar.gz

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: local-coding-agent
-Version: 1.1.0
+Version: 1.1.1
 Summary: Harness-agnostic controller delegating atomic coding tasks to local Ollama & llama.cpp models via MCP, Agent Skills, and direct CLI.
 Author: Local Coding Agent Contributors
 License: MIT
@@ -15,6 +15,7 @@ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
+Requires-Dist: mcp==2.0.0
 Provides-Extra: mcp
 Requires-Dist: mcp==2.0.0; extra == "mcp"
 Provides-Extra: dev
@@ -25,10 +26,12 @@ Dynamic: license-file
 
 # Local Coding Agent
 
+<!-- mcp-name: io.github.pvnc228/local-coding-agent -->
+
 Local execution and safety layer for delegating coding tasks from cloud agents to models on your own machine.
 
 [![CI](https://github.com/pvnc228/local-coding-agent/actions/workflows/ci.yml/badge.svg)](https://github.com/pvnc228/local-coding-agent/actions/workflows/ci.yml)
-[![Version](https://img.shields.io/badge/version-1.1.0-blue.svg)](pyproject.toml)
+[![Version](https://img.shields.io/badge/version-1.1.1-blue.svg)](pyproject.toml)
 [![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
 [![License](https://img.shields.io/badge/license-MIT-green.svg)](LICENSE)
 
@@ -53,13 +56,13 @@ The same gate backs `apply_proposal`, `local-agent apply`, and the desktop apply
 
 ## Choose your download
 
-### Current public release: v1.0.4
+### Current public release: v1.1.0
 
 Choose the CLI release for terminal workflows, scripts, CI, or integration with another AI coding environment. It provides the `local-agent` command, MCP server, Agent Skill, diagnostics, sessions, model tools, and task delegation.
 
-[Download the CLI wheel](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/local_coding_agent-1.0.4-py3-none-any.whl) or [CLI source archive](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/local_coding_agent-1.0.4.tar.gz). Windows users can [download the Desktop Harness installer](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/Local.AI.Coding.Harness_1.0.4_x64-setup.exe) and verify the published files with [SHA256SUMS.txt](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/SHA256SUMS.txt).
+[Download the CLI wheel](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/local_coding_agent-1.1.0-py3-none-any.whl) or [CLI source archive](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/local_coding_agent-1.1.0.tar.gz). Windows users can [download the Desktop Harness installer](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/Local.AI.Coding.Harness_1.1.0_x64-setup.exe) and verify the published files with [SHA256SUMS.txt](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/SHA256SUMS.txt).
 
-The `v1.0.4` release contains the CLI archives and the Windows Desktop Harness installer. Local inference remains local; install [Ollama](https://ollama.com/) or run a compatible local model server before requesting live model responses.
+The `v1.1.0` release contains the CLI archives and the Windows Desktop Harness installer. Local inference remains local; install [Ollama](https://ollama.com/) or run a compatible local model server before requesting live model responses.
 
 ## Desktop Harness
 
@@ -91,7 +94,7 @@ Requirements:
 Install the CLI package:
 
 ```bash
-pipx install "local-coding-agent[mcp]"
+pipx install local-coding-agent
 ```
 
 Or install from a checkout:
@@ -99,7 +102,7 @@ Or install from a checkout:
 ```bash
 git clone https://github.com/pvnc228/local-coding-agent.git
 cd local-coding-agent
-python -m pip install -e ".[mcp]"
+python -m pip install -e .
 ```
 
 Check the installation:

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/README.md

@@ -1,9 +1,11 @@
 # Local Coding Agent
 
+<!-- mcp-name: io.github.pvnc228/local-coding-agent -->
+
 Local execution and safety layer for delegating coding tasks from cloud agents to models on your own machine.
 
 [![CI](https://github.com/pvnc228/local-coding-agent/actions/workflows/ci.yml/badge.svg)](https://github.com/pvnc228/local-coding-agent/actions/workflows/ci.yml)
-[![Version](https://img.shields.io/badge/version-1.1.0-blue.svg)](pyproject.toml)
+[![Version](https://img.shields.io/badge/version-1.1.1-blue.svg)](pyproject.toml)
 [![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
 [![License](https://img.shields.io/badge/license-MIT-green.svg)](LICENSE)
 
@@ -28,13 +30,13 @@ The same gate backs `apply_proposal`, `local-agent apply`, and the desktop apply
 
 ## Choose your download
 
-### Current public release: v1.0.4
+### Current public release: v1.1.0
 
 Choose the CLI release for terminal workflows, scripts, CI, or integration with another AI coding environment. It provides the `local-agent` command, MCP server, Agent Skill, diagnostics, sessions, model tools, and task delegation.
 
-[Download the CLI wheel](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/local_coding_agent-1.0.4-py3-none-any.whl) or [CLI source archive](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/local_coding_agent-1.0.4.tar.gz). Windows users can [download the Desktop Harness installer](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/Local.AI.Coding.Harness_1.0.4_x64-setup.exe) and verify the published files with [SHA256SUMS.txt](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/SHA256SUMS.txt).
+[Download the CLI wheel](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/local_coding_agent-1.1.0-py3-none-any.whl) or [CLI source archive](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/local_coding_agent-1.1.0.tar.gz). Windows users can [download the Desktop Harness installer](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/Local.AI.Coding.Harness_1.1.0_x64-setup.exe) and verify the published files with [SHA256SUMS.txt](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/SHA256SUMS.txt).
 
-The `v1.0.4` release contains the CLI archives and the Windows Desktop Harness installer. Local inference remains local; install [Ollama](https://ollama.com/) or run a compatible local model server before requesting live model responses.
+The `v1.1.0` release contains the CLI archives and the Windows Desktop Harness installer. Local inference remains local; install [Ollama](https://ollama.com/) or run a compatible local model server before requesting live model responses.
 
 ## Desktop Harness
 
@@ -66,7 +68,7 @@ Requirements:
 Install the CLI package:
 
 ```bash
-pipx install "local-coding-agent[mcp]"
+pipx install local-coding-agent
 ```
 
 Or install from a checkout:
@@ -74,7 +76,7 @@ Or install from a checkout:
 ```bash
 git clone https://github.com/pvnc228/local-coding-agent.git
 cd local-coding-agent
-python -m pip install -e ".[mcp]"
+python -m pip install -e .
 ```
 
 Check the installation:

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent/__init__.py

@@ -1,6 +1,6 @@
 """Bounded controller components for delegating small coding tasks."""
 
-__version__ = "1.1.0"
+__version__ = "1.1.1"
 
 
 from .acp_server import (

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent/cli/_handlers2.py

@@ -525,7 +525,11 @@ def _handle_serve_mcp(args: argparse.Namespace) -> int:
     from ..service import DelegationService
 
     ws_path = str(Path(args.workspace).resolve())
-    service = DelegationService({args.workspace_ref: ws_path})
+    service = DelegationService(
+        {args.workspace_ref: ws_path},
+        default_profile=args.profile,
+        endpoint=args.endpoint,
+    )
     server = build_server(service, enable_tasks=args.enable_tasks)
     server.run(transport="stdio")
     return 0

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent/mcp_server.py

@@ -33,7 +33,7 @@ except ImportError:  # pragma: no cover - pydantic comes with mcp
 
 _CALLER_ID = "mcp-stdio"
 _SERVER_NAME = "local-coding-agent"
-_SERVER_VERSION = "1.1.0"
+_SERVER_VERSION = "1.1.1"
 
 
 
@@ -205,7 +205,7 @@ def build_server(
             execution_gate=runtime_gate,
             task_store=task_store,
         )
-        extensions.append(TasksExtension(pool, caller_id=_CALLER_ID))
+        extensions.append(TasksExtension(pool, caller_id=_CALLER_ID, default_profile=service.default_profile))
 
 
     server = MCPServer(
@@ -268,10 +268,11 @@ def build_server(
     async def delegate_code(
         request_id: str,
         workspace_ref: str,
-        model_profile: str,
         task: dict[str, Any],
+        model_profile: str | None = None,
     ) -> CallToolResult:
-        request = _delegate_request(request_id, workspace_ref, model_profile, task)
+        # An omitted profile falls back to the one the host started the server with (--profile).
+        request = _delegate_request(request_id, workspace_ref, model_profile or service.default_profile, task)
         try:
             result = await execution_gate.run(service.delegate, _CALLER_ID, request)
         except ExecutionOverload:

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent/service.py

@@ -83,6 +83,8 @@ class DelegationService:
         max_turns: int = 4,
         max_cached_results: int = 256,
         preflight_budget: TaskBudget = TaskBudget(),
+        default_profile: str | None = None,
+        endpoint: str | None = None,
     ) -> None:
         if max_turns <= 0:
             raise ValueError("max_turns must be positive")
@@ -104,6 +106,8 @@ class DelegationService:
         self._max_turns = max_turns
         self._max_cached_results = max_cached_results
         self._preflight_budget = preflight_budget
+        self.default_profile = default_profile
+        self.endpoint = endpoint
         self._cache: OrderedDict[tuple[str, str, str], _CachedResult] = OrderedDict()
         self._cache_lock = RLock()
 
@@ -257,7 +261,7 @@ class DelegationService:
         if not report.accepted:
             return self._policy_failure("preflight_rejected", report.reason or "preflight_rejected")
         try:
-            profile = get_profile(request.model_profile)
+            profile = get_profile(request.model_profile, **({"endpoint": self.endpoint} if self.endpoint else {}))
         except ValueError:
             return self._policy_failure(
                 "unknown_model_profile",

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent/skill_config.py

@@ -56,7 +56,7 @@ Delegates one atomic, proposal-only coding task to a local Ollama model.
 #### Parameters:
 - `request_id` *(string, required)*: A unique caller-generated idempotency key (e.g. `task-fix-auth-01`).
 - `workspace_ref` *(string, required)*: The registered workspace identifier (usually `"workspace"`).
-- `model_profile` *(string, required)*: Named profile (e.g. `"qwen2.5-coder"`, `"qwen3-8b-q6k"`, `"devstral-small-2-24b"`).
+- `model_profile` *(string, optional when the server was started with `--profile`)*: Named profile (e.g. `"qwen2.5-coder"`, `"qwen3-8b-q6k"`, `"devstral-small-2-24b"`).
 - `task` *(object, required)*: The structured task envelope:
   - `id` *(string)*: Short descriptive identifier (e.g. `"fix-null-dereference"`).
   - `goal` *(string)*: 1 concise sentence describing the goal.

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent/tasks.py

@@ -147,9 +147,11 @@ if _MCP_AVAILABLE:
             caller_id: str,
             ttl_ms: int = _DEFAULT_TTL_MS,
             poll_interval_ms: int = _DEFAULT_POLL_INTERVAL_MS,
+            default_profile: str | None = None,
         ) -> None:
             self._pool = pool
             self._caller_id = caller_id
+            self._default_profile = default_profile
             self._ttl_ms = ttl_ms
             self._poll_interval_ms = poll_interval_ms
 
@@ -185,7 +187,7 @@ if _MCP_AVAILABLE:
                 request = DelegationRequest(
                     request_id=arguments.get("request_id"),
                     workspace_ref=arguments.get("workspace_ref"),
-                    model_profile=arguments.get("model_profile"),
+                    model_profile=arguments.get("model_profile") or self._default_profile,
                     task=TaskEnvelope.from_mapping(arguments.get("task") or {}),
                 )
             except (TypeError, ValueError):

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: local-coding-agent
-Version: 1.1.0
+Version: 1.1.1
 Summary: Harness-agnostic controller delegating atomic coding tasks to local Ollama & llama.cpp models via MCP, Agent Skills, and direct CLI.
 Author: Local Coding Agent Contributors
 License: MIT
@@ -15,6 +15,7 @@ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
+Requires-Dist: mcp==2.0.0
 Provides-Extra: mcp
 Requires-Dist: mcp==2.0.0; extra == "mcp"
 Provides-Extra: dev
@@ -25,10 +26,12 @@ Dynamic: license-file
 
 # Local Coding Agent
 
+<!-- mcp-name: io.github.pvnc228/local-coding-agent -->
+
 Local execution and safety layer for delegating coding tasks from cloud agents to models on your own machine.
 
 [![CI](https://github.com/pvnc228/local-coding-agent/actions/workflows/ci.yml/badge.svg)](https://github.com/pvnc228/local-coding-agent/actions/workflows/ci.yml)
-[![Version](https://img.shields.io/badge/version-1.1.0-blue.svg)](pyproject.toml)
+[![Version](https://img.shields.io/badge/version-1.1.1-blue.svg)](pyproject.toml)
 [![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
 [![License](https://img.shields.io/badge/license-MIT-green.svg)](LICENSE)
 
@@ -53,13 +56,13 @@ The same gate backs `apply_proposal`, `local-agent apply`, and the desktop apply
 
 ## Choose your download
 
-### Current public release: v1.0.4
+### Current public release: v1.1.0
 
 Choose the CLI release for terminal workflows, scripts, CI, or integration with another AI coding environment. It provides the `local-agent` command, MCP server, Agent Skill, diagnostics, sessions, model tools, and task delegation.
 
-[Download the CLI wheel](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/local_coding_agent-1.0.4-py3-none-any.whl) or [CLI source archive](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/local_coding_agent-1.0.4.tar.gz). Windows users can [download the Desktop Harness installer](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/Local.AI.Coding.Harness_1.0.4_x64-setup.exe) and verify the published files with [SHA256SUMS.txt](https://github.com/pvnc228/local-coding-agent/releases/download/v1.0.4/SHA256SUMS.txt).
+[Download the CLI wheel](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/local_coding_agent-1.1.0-py3-none-any.whl) or [CLI source archive](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/local_coding_agent-1.1.0.tar.gz). Windows users can [download the Desktop Harness installer](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/Local.AI.Coding.Harness_1.1.0_x64-setup.exe) and verify the published files with [SHA256SUMS.txt](https://github.com/pvnc228/local-coding-agent/releases/download/v1.1.0/SHA256SUMS.txt).
 
-The `v1.0.4` release contains the CLI archives and the Windows Desktop Harness installer. Local inference remains local; install [Ollama](https://ollama.com/) or run a compatible local model server before requesting live model responses.
+The `v1.1.0` release contains the CLI archives and the Windows Desktop Harness installer. Local inference remains local; install [Ollama](https://ollama.com/) or run a compatible local model server before requesting live model responses.
 
 ## Desktop Harness
 
@@ -91,7 +94,7 @@ Requirements:
 Install the CLI package:
 
 ```bash
-pipx install "local-coding-agent[mcp]"
+pipx install local-coding-agent
 ```
 
 Or install from a checkout:
@@ -99,7 +102,7 @@ Or install from a checkout:
 ```bash
 git clone https://github.com/pvnc228/local-coding-agent.git
 cd local-coding-agent
-python -m pip install -e ".[mcp]"
+python -m pip install -e .
 ```
 
 Check the installation:

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/local_coding_agent.egg-info/requires.txt

@@ -1,3 +1,4 @@
+mcp==2.0.0
 
 [desktop]
 pywebview>=5.0.0

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "local-coding-agent"
-version = "1.1.0"
+version = "1.1.1"
 
 description = "Harness-agnostic controller delegating atomic coding tasks to local Ollama & llama.cpp models via MCP, Agent Skills, and direct CLI."
 
@@ -22,12 +22,14 @@ classifiers = [
     "Topic :: Software Development :: Code Generators",
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
 ]
+dependencies = ["mcp==2.0.0"]
 
 [project.scripts]
 local-agent = "local_coding_agent.cli:main"
 local-coding-agent = "local_coding_agent.cli:main"
 
 [project.optional-dependencies]
+# Kept so existing `local-coding-agent[mcp]` installs keep resolving; the SDK is a core dependency.
 mcp = ["mcp==2.0.0"]
 dev = ["pytest"]
 desktop = ["pywebview>=5.0.0"]

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/tests/test_cli.py

@@ -108,6 +108,21 @@ class CliTests(unittest.TestCase):
         self.assertEqual(args.workspace, "c:/code")
         self.assertTrue(args.enable_tasks)
 
+    def test_serve_mcp_passes_profile_and_endpoint_to_service(self):
+        from local_coding_agent.cli._handlers2 import _handle_serve_mcp
+
+        with tempfile.TemporaryDirectory() as temp_dir:
+            args = build_parser().parse_args(
+                ["serve-mcp", "--workspace", temp_dir, "--profile", "qwen3-8b-q6k", "--endpoint", "http://127.0.0.1:8080"]
+            )
+            with mock.patch("local_coding_agent.mcp_server.build_server") as build_server:
+                self.assertEqual(_handle_serve_mcp(args), 0)
+
+        service = build_server.call_args.args[0]
+        self.assertEqual(service.default_profile, "qwen3-8b-q6k")
+        self.assertEqual(service.endpoint, "http://127.0.0.1:8080")
+        build_server.return_value.run.assert_called_once_with(transport="stdio")
+
     def test_cli_subcommand_monitor(self):
         args = build_parser().parse_args(["monitor", "--port", "9000"])
         self.assertEqual(args.subcommand, "monitor")

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/tests/test_mcp_server.py

@@ -81,6 +81,32 @@ class McpSdkServerTests(unittest.TestCase):
         self.assertEqual(result.structured_content["status"], "accepted")
         self.assertFalse(result.structured_content.get("applied", False))
 
+    def test_delegate_code_uses_server_default_profile_when_omitted(self):
+        profiles = []
+        with tempfile.TemporaryDirectory() as temp_dir:
+            workspace = Path(temp_dir)
+            (workspace / "allowed.py").write_text("VALUE = 1\n", encoding="utf-8")
+            service = DelegationService(
+                {"fixture": workspace},
+                model_factory=lambda profile: profiles.append(profile.name) or FakeModel(),
+                default_profile="qwen2.5-1.5b",
+            )
+            server = build_server(service)
+            args = self._arguments()
+            del args["model_profile"]
+
+            async def run():
+                from mcp.client import Client
+
+                async with Client(server) as client:
+                    return await client.call_tool("delegate_code", args)
+
+            result = asyncio.run(run())
+
+        self.assertFalse(result.is_error)
+        self.assertEqual(result.structured_content["status"], "accepted")
+        self.assertEqual(profiles, ["qwen2.5-1.5b"])
+
     def test_delegate_code_unknown_workspace_is_error(self):
         with tempfile.TemporaryDirectory() as temp_dir:
             server = build_server(self._service(Path(temp_dir)))

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/tests/test_service.py

@@ -126,6 +126,23 @@ class DelegationServiceTests(unittest.TestCase):
         self.assertEqual(unknown_profile["error"]["kind"], "unknown_model_profile")
         self.assertEqual(calls, [])
 
+    def test_endpoint_override_applies_to_resolved_profile(self):
+        profiles = []
+        with tempfile.TemporaryDirectory() as temp_dir:
+            workspace = Path(temp_dir)
+            (workspace / "allowed.py").write_text("VALUE = 1\n", encoding="utf-8")
+            model = FakeModel(self._candidate())
+            service = DelegationService(
+                {"fixture": workspace},
+                model_factory=lambda profile: profiles.append(profile) or model,
+                endpoint="http://127.0.0.1:8080",
+            )
+
+            result = service.delegate("caller-a", self._request())
+
+        self.assertEqual(result["status"], "accepted")
+        self.assertEqual([profile.endpoint for profile in profiles], ["http://127.0.0.1:8080"])
+
     def test_rejects_reused_request_id_with_a_different_payload(self):
         with tempfile.TemporaryDirectory() as temp_dir:
             workspace = Path(temp_dir)

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/tests/test_tasks.py

@@ -106,6 +106,8 @@ class FailingModel:
 
 
 class MixedBlockingService:
+    default_profile = None
+
     def __init__(self):
         self.started = Event()
         self.overlap = Event()

{local_coding_agent-1.1.0 → local_coding_agent-1.1.1}/tests/test_version_consistency.py

@@ -10,7 +10,7 @@ from pathlib import Path
 
 
 ROOT = Path(__file__).resolve().parents[1]
-PUBLIC_RELEASE_VERSION = "1.0.4"
+PUBLIC_RELEASE_VERSION = "1.1.0"
 
 
 def _project_version() -> str:
@@ -54,6 +54,27 @@ def test_all_release_metadata_matches_pyproject_version():
     )
 
 
+def test_mcp_registry_metadata_matches_the_pypi_package():
+    expected = _project_version()
+    server = json.loads((ROOT / "server.json").read_text(encoding="utf-8-sig"))
+    pyproject = (ROOT / "pyproject.toml").read_text(encoding="utf-8-sig")
+    readme = (ROOT / "README.md").read_text(encoding="utf-8-sig")
+    (package,) = server["packages"]
+
+    assert server["name"] == "io.github.pvnc228/local-coding-agent"
+    assert len(server["description"]) <= 100
+    assert server["version"] == expected
+    assert package["registryType"] == "pypi"
+    assert package["identifier"] == re.search(r'^name\s*=\s*"([^"]+)"', pyproject, re.MULTILINE).group(1)
+    assert package["version"] == expected
+    assert package["transport"] == {"type": "stdio"}
+    # The registry verifies PyPI ownership through this marker in the published README.
+    assert re.search(rf"mcp-name: {re.escape(server['name'])}(?:\s|-->)", readme)
+    # Registry clients launch the bare package, so the MCP SDK cannot be an optional extra.
+    dependencies = re.search(r"^dependencies\s*=\s*\[(.*?)\]", pyproject, re.MULTILINE | re.DOTALL)
+    assert dependencies and '"mcp==' in dependencies.group(1)
+
+
 def test_ci_and_release_workflows_gate_and_publish_the_windows_installer():
     ci = (ROOT / ".github" / "workflows" / "ci.yml").read_text(encoding="utf-8-sig")
     release = (ROOT / ".github" / "workflows" / "release.yml").read_text(encoding="utf-8-sig")