test(mcp): smoke the okf 1.1 server through the run's own MCP path -- works by full path, not by bare name

A new test builds a small invented collection with `okf build` and connects through
load_mcp_config -> build_mcp_tools -> async with. No model call, no network. It measures:
- the four tools are offered, and a narrower allowed_tools narrows them;
- okf_list names the collection;
- okf_ask returns the excerpt.

That holds only when the config gives the full path to okf 1.1. With the README's bare `okf`,
a run under `uv run` or an activated venv finds this project's own dependency
llm-ingestion-okf 0.8.5 first. That version has no `mcp` subcommand, and the server fails at
startup ("invalid choice: 'mcp'"). This is pinned as xfail(strict=True, raises=ToolException)
and not repaired, per the order.

The tests skip when no okf >= 1.1.0 is found on PATH outside the venv. The version is measured
from the server's own initialize answer. README limits now state what was measured. The
published-surface pin moves from 518 to 519 for the new file.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

README.md

@@ -390,9 +390,13 @@ With it, the server and each allowed tool are named in the announcement before t
 
 **Limits, stated plainly.**
 
-- The link between this framework and the `okf` server has **not been tried in a real run**. Only
+- **Measured without a model** (`tests/test_okf_server_smoke.py`, against a small invented
-  the config file is validated (`tests/test_okf_server_example.py`, against the same loader a run
+  collection): through the same loader and tool builder a run uses, `okf` 1.1 offers exactly the
-  uses).
+  four tools, `allowed_tools` narrows what is offered, `okf_list` names the collection and one
+  `okf_ask` returns the excerpt. **This works only when `command` is the full path to the `okf`
+  1.1 executable.** With the bare `okf` shown above, a run started with `uv run` or from an
+  activated environment finds this framework's own `okf` dependency (0.8.5) first. That version
+  has no `mcp` command, and the server fails at startup (`invalid choice: 'mcp'`).
 - It is **not measured** whether the agents follow the way of working the server describes.
 - `--prepass-payload` does not accept the v1.1 format (`okf-consumption/2`). That is a known gap.
 

tests/test_b_gate.py

@@ -601,7 +601,7 @@ _WRITERS_DEFINED = 10
 _WRITERS_IN_RUN_PATH = 7
 #: publiserte filer i git-manifestet (uttrekk og arbeidstre gir SAMME tall — det var hele poenget
 #: med å slutte å telle filtreet: 435 i arbeidstreet var to gitignorerte .local.md-filer).
-_PUBLISHED_TODAY = 518
+_PUBLISHED_TODAY = 519
 _UNDECODABLE_TODAY = 1
 
 

tests/test_okf_server_smoke.py

@@ -0,0 +1,171 @@
+"""Smoke test: does a run's own MCP path actually reach the ``okf`` v1.1 server?
+
+The README documents ``examples/okf-server.mcp.json``; ``test_okf_server_example.py`` only proves
+the file loads. These tests connect through the same two functions a run uses
+(``load_mcp_config`` -> ``build_mcp_tools``) and the same ``async with`` a run enters, against a
+tiny invented collection built in ``tmp_path``. No model call, no network.
+
+They are skipped, with the reason named, when no ``okf`` >= 1.1.0 is found on ``PATH`` outside
+this environment's own ``bin``. The version is MEASURED from the server's own ``initialize``
+answer (``serverInfo.version``), never assumed from the name.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+from typing import Any
+
+import pytest
+from agent_framework.exceptions import ToolException
+
+from portfolio_optimiser.mcp_tools import build_mcp_tools, load_mcp_config
+
+FOUR = ["okf_list", "okf_describe", "okf_ask", "okf_fetch"]
+VENV_BIN = str(Path(sys.executable).parent)
+
+_INIT = json.dumps(
+    {
+        "jsonrpc": "2.0",
+        "id": 1,
+        "method": "initialize",
+        "params": {
+            "protocolVersion": "2025-06-18",
+            "capabilities": {},
+            "clientInfo": {"name": "smoke", "version": "0"},
+        },
+    }
+)
+
+
+def _server_version(exe: str, root: Path) -> tuple[int, ...] | None:
+    """The version the server reports about itself, or None if ``exe`` serves no MCP at all."""
+    try:
+        done = subprocess.run(
+            [exe, "mcp", "--root", str(root)],
+            input=_INIT + "\n",
+            capture_output=True,
+            text=True,
+            timeout=60,
+        )
+        version = json.loads(done.stdout.splitlines()[0])["result"]["serverInfo"]["version"]
+        return tuple(int(part) for part in version.split(".")[:3])
+    except (IndexError, KeyError, ValueError, OSError, subprocess.TimeoutExpired):
+        return None
+
+
+@pytest.fixture
+def okf_1_1(tmp_path: Path) -> str:
+    """Absolute path of an ``okf`` >= 1.1.0 on PATH, looked up OUTSIDE this environment's bin."""
+    outside = os.pathsep.join(
+        p for p in os.environ.get("PATH", "").split(os.pathsep) if p and p != VENV_BIN
+    )
+    exe = shutil.which("okf", path=outside)
+    if exe is None:
+        pytest.skip(
+            "no `okf` on PATH outside this environment -- the smoke test needs okf >= 1.1.0"
+        )
+    probe_root = tmp_path / "empty-root"
+    probe_root.mkdir()
+    version = _server_version(exe, probe_root)
+    if version is None or version < (1, 1, 0):
+        pytest.skip(f"`{exe}` reports version {version} over MCP -- the smoke test needs >= 1.1.0")
+    return exe
+
+
+@pytest.fixture
+def invented_root(tmp_path: Path, okf_1_1: str) -> Path:
+    """A two-document collection about an invented island, built with ``okf build``."""
+    inbox = tmp_path / "inbox"
+    inbox.mkdir()
+    (inbox / "lighthouse.md").write_text(
+        "# Lighthouse keeping\n\n## Lamp maintenance\n\n"
+        "The quartz lamp at Vardholm point is polished every Tuesday with walrus wax.\n",
+        encoding="utf-8",
+    )
+    (inbox / "ferry.md").write_text(
+        "# Ferry timetable\n\n## Winter crossings\n\n"
+        "The Vardholm ferry sails twice a day in winter, leaving at dawn and dusk.\n",
+        encoding="utf-8",
+    )
+    root = tmp_path / "root"
+    subprocess.run(
+        [okf_1_1, "build", str(inbox), "--bundle", str(root / "vardholm"), "--gate", "none"]
+        + ["--bundle-id", "vardholm", "--okf-version", "1.0"],
+        check=True,
+        capture_output=True,
+        timeout=120,
+    )
+    return root
+
+
+def _config(tmp_path: Path, command: str, root: Path, allowed: list[str]) -> Path:
+    path = tmp_path / "okf.mcp.json"
+    server = {
+        "name": "okf",
+        "transport": "stdio",
+        "command": command,
+        "args": ["mcp", "--root", str(root)],
+        "allowed_tools": allowed,
+        "timeout_seconds": 60,
+    }
+    path.write_text(json.dumps({"servers": [server]}), encoding="utf-8")
+    return path
+
+
+def _text(result: Any) -> str:
+    return "\n".join(getattr(item, "text", None) or "" for item in result)
+
+
+async def test_okf_1_1_by_full_path_offers_four_tools_lists_and_answers(
+    tmp_path: Path, okf_1_1: str, invented_root: Path
+) -> None:
+    (tool,) = build_mcp_tools(load_mcp_config(_config(tmp_path, okf_1_1, invented_root, FOUR)))
+    async with tool:
+        # (a) the four are offered
+        assert sorted(f.name for f in tool.functions) == sorted(FOUR)
+        # (b) okf_list names the invented collection
+        listed = json.loads((await tool.call_tool("okf_list"))[0].text)
+        assert [b["bundle_id"] for b in listed["bundles"]] == ["vardholm"]
+        # (c) one okf_ask in the fixture's words delivers the excerpt
+        asked = _text(
+            await tool.call_tool(
+                "okf_ask", bundle="vardholm", questions=["How is the quartz lamp polished?"]
+            )
+        )
+        assert "walrus wax" in asked
+
+
+async def test_allowed_tools_is_load_bearing_against_the_real_server(
+    tmp_path: Path, okf_1_1: str, invented_root: Path
+) -> None:
+    """The server offers exactly the four, so (a) alone cannot show the allowlist filters.
+    A narrower list must narrow what the agents are offered."""
+    config = _config(tmp_path, okf_1_1, invented_root, ["okf_list", "okf_ask"])
+    (tool,) = build_mcp_tools(load_mcp_config(config))
+    async with tool:
+        assert sorted(f.name for f in tool.functions) == ["okf_ask", "okf_list"]
+
+
+@pytest.mark.xfail(
+    strict=True,
+    raises=ToolException,
+    reason=(
+        "MEASURED 2026-09-21: with the README's bare `okf` command and this environment's bin "
+        "first on PATH (as under `uv run` or an activated venv), the name resolves to the `okf` "
+        "script of po's own dependency llm-ingestion-okf 0.8.5, which has no `mcp` subcommand: "
+        "stderr `okf: error: argument command: invalid choice: 'mcp'`, and MAF raises "
+        "ToolException '... failed to initialize: Connection closed' on `async with`."
+    ),
+)
+async def test_readme_bare_okf_command_reaches_the_1_1_server(
+    tmp_path: Path, okf_1_1: str, invented_root: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    monkeypatch.setenv("PATH", VENV_BIN + os.pathsep + os.environ.get("PATH", ""))
+    (tool,) = build_mcp_tools(load_mcp_config(_config(tmp_path, "okf", invented_root, FOUR)))
+    async with tool:
+        assert sorted(f.name for f in tool.functions) == sorted(FOUR)