Improve MCP config auto discovery

Author: ithiria894

README.md

@@ -40,7 +40,9 @@ python3 -m venv .venv-intentprobe
 .venv-intentprobe/bin/intentprobe scan-config auto --format summary
 ```
 
-Then scan the MCP tools already configured on your machine:
+Then scan the MCP tools already configured on your machine. `scan-config auto`
+checks common Claude Desktop, Claude Code, Codex, Cursor, Windsurf, and repo MCP
+config locations:
 
 ```bash
 intentprobe scan-config auto --format summary
@@ -175,7 +177,7 @@ Same test sets. Same split. Same seed. Every number is reproducible from `resear
 ## Try it
 
 ```bash
-# Scan Claude/Cursor/Claude Code MCP configs already on this machine
+# Scan Claude/Cursor/Codex MCP configs already on this machine
 intentprobe scan-config auto --format summary
 
 # Scan a suspicious tool description
@@ -300,8 +302,10 @@ blocked.
 
   scan-config:
   ├── Claude Desktop           claude_desktop_config.json
-  ├── Claude Code              ~/.claude/mcp.json
+  ├── Claude Code              ~/.claude.json, ~/.claude/mcp.json
+  ├── Codex                    ~/.codex/config.toml
   ├── Cursor                   ~/.cursor/mcp.json
+  ├── Windsurf                 ~/.codeium/windsurf/mcp_config.json
   └── local repo               .mcp.json
 
   runtime:

intentprobe/__init__.py

@@ -1,3 +1,3 @@
 """intentprobe public package."""
 
-__version__ = "0.1.2"
+__version__ = "0.1.3"

intentprobe/scanner/configs.py

@@ -8,6 +8,11 @@
 from pathlib import Path
 from typing import Any
 
+try:
+    import tomllib
+except ModuleNotFoundError:  # pragma: no cover - Python 3.10 fallback
+    import tomli as tomllib
+
 from .core import CORE_VERSION, DECISION_POLICY_NAME
 from .hook import ScanSubject, max_decision, object_subject
 
@@ -20,13 +25,9 @@
 }
 
 REVIEW_FLAG_IDS = {
-    "env-secrets",
-    "remote-http",
     "browser-access",
     "filesystem-access",
     "code-execution",
-    "email-or-identity",
-    "repo-or-ticketing-access",
 }
 
 
@@ -57,7 +58,9 @@ def default_config_candidates(cwd: Path | None = None) -> list[ConfigCandidate]:
     working_dir = cwd or Path.cwd()
     candidates = [
         ConfigCandidate("Claude Desktop", home / "Library/Application Support/Claude/claude_desktop_config.json"),
-        ConfigCandidate("Claude Code", home / ".claude/mcp.json"),
+        ConfigCandidate("Claude Code Global", home / ".claude.json"),
+        ConfigCandidate("Claude Code MCP", home / ".claude/mcp.json"),
+        ConfigCandidate("Codex", home / ".codex/config.toml"),
         ConfigCandidate("Cursor", home / ".cursor/mcp.json"),
         ConfigCandidate("Cursor User", home / "Library/Application Support/Cursor/User/mcp.json"),
         ConfigCandidate("Windsurf", home / ".codeium/windsurf/mcp_config.json"),
@@ -85,19 +88,29 @@ def config_candidates_from_target(target: str | Path | None, cwd: Path | None =
     return [ConfigCandidate("custom", Path(target).expanduser())]
 
 
-def load_json_config(path: Path, max_file_bytes: int) -> tuple[dict[str, Any] | None, str | None]:
+def load_structured_config(path: Path, max_file_bytes: int) -> tuple[dict[str, Any] | None, str | None]:
     try:
         raw = path.read_bytes()
     except OSError as exc:
         return None, f"read_error: {exc}"
     if len(raw) > max_file_bytes:
         return None, f"file_too_large: {len(raw)} bytes > {max_file_bytes}"
     try:
-        payload = json.loads(raw.decode("utf-8"))
+        text = raw.decode("utf-8")
     except UnicodeDecodeError as exc:
         return None, f"decode_error: {exc}"
-    except json.JSONDecodeError as exc:
-        return None, f"json_error: {exc}"
+
+    if path.suffix == ".toml":
+        try:
+            payload = tomllib.loads(text)
+        except tomllib.TOMLDecodeError as exc:
+            return None, f"toml_error: {exc}"
+    else:
+        try:
+            payload = json.loads(text)
+        except json.JSONDecodeError as exc:
+            return None, f"json_error: {exc}"
+
     if not isinstance(payload, dict):
         return None, f"unsupported_json_type: {type(payload).__name__}"
     return payload, None
@@ -147,7 +160,7 @@ def collect_config_servers(
             configs.append(public_config)
             continue
 
-        payload, error = load_json_config(path, max_file_bytes)
+        payload, error = load_structured_config(path, max_file_bytes)
         if error is not None or payload is None:
             public_config.update({"status": "invalid", "error": error})
             configs.append(public_config)
@@ -194,7 +207,7 @@ def inventory_flags(server: ConfigServer) -> list[dict[str, str]]:
         flags.append(
             {
                 "id": "env-secrets",
-                "level": "review",
+                "level": "info",
                 "reason": "server config includes environment variables; values are redacted",
             }
         )
@@ -203,7 +216,7 @@ def inventory_flags(server: ConfigServer) -> list[dict[str, str]]:
         flags.append(
             {
                 "id": "remote-http",
-                "level": "review",
+                "level": "info",
                 "reason": "server connects to a remote MCP endpoint",
             }
         )
@@ -248,7 +261,7 @@ def inventory_flags(server: ConfigServer) -> list[dict[str, str]]:
         flags.append(
             {
                 "id": "email-or-identity",
-                "level": "review",
+                "level": "info",
                 "reason": "server appears connected to email, calendar, OAuth, or identity data",
             }
         )
@@ -257,7 +270,7 @@ def inventory_flags(server: ConfigServer) -> list[dict[str, str]]:
         flags.append(
             {
                 "id": "repo-or-ticketing-access",
-                "level": "review",
+                "level": "info",
                 "reason": "server appears connected to code repositories or ticketing systems",
             }
         )
@@ -276,7 +289,7 @@ def product_decision_for_config_scan(risk: dict[str, Any], flags: list[dict[str,
 
     scanner_decision = str(risk.get("decision", "allow"))
     static_score = float(risk.get("static_score") or 0.0)
-    review_flags = [flag for flag in flags if flag.get("id") in REVIEW_FLAG_IDS]
+    review_flags = [flag for flag in flags if flag.get("level") == "review" and flag.get("id") in REVIEW_FLAG_IDS]
     reasons: list[str] = []
 
     if scanner_decision == "block":

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "intentprobe"
-version = "0.1.2"
+version = "0.1.3"
 description = "Local activation-probe scanner for AI agent tools, MCP servers, and skills."
 readme = "README.md"
 requires-python = ">=3.10,<3.14"
@@ -17,6 +17,7 @@ dependencies = [
   "psutil",
   "scikit-learn",
   "sentencepiece",
+  "tomli; python_version < '3.11'",
   "torch",
   "transformers>=4.40",
 ]

tests/test_scan_config.py

@@ -10,6 +10,7 @@
     ConfigServer,
     build_config_scan_payload,
     collect_config_servers,
+    default_config_candidates,
     inventory_flags,
     product_decision_for_config_scan,
     subject_for_server,
@@ -31,6 +32,41 @@ def test_config_without_mcp_servers_is_skipped(self) -> None:
         self.assertEqual("no_mcp_servers", configs[0]["status"])
         self.assertEqual(0, configs[0]["server_count"])
 
+    def test_toml_codex_mcp_servers_are_collected(self) -> None:
+        with tempfile.TemporaryDirectory() as tmp:
+            path = Path(tmp) / "config.toml"
+            path.write_text(
+                "\n".join(
+                    [
+                        "[mcp_servers.github]",
+                        'command = "npx"',
+                        'args = ["-y", "@modelcontextprotocol/server-github"]',
+                        "",
+                        "[mcp_servers.chrome]",
+                        'command = "npx"',
+                        'args = ["-y", "chrome-devtools-mcp"]',
+                    ]
+                )
+            )
+
+            configs, servers = collect_config_servers(
+                [ConfigCandidate("Codex", path)],
+                max_file_bytes=200_000,
+            )
+
+        self.assertEqual("scanned", configs[0]["status"])
+        self.assertEqual("mcp_servers", configs[0]["servers_key"])
+        self.assertEqual(2, configs[0]["server_count"])
+        self.assertEqual(["chrome", "github"], [server.name for server in servers])
+
+    def test_auto_candidates_include_claude_global_and_codex(self) -> None:
+        sources = {candidate.source: str(candidate.path) for candidate in default_config_candidates(Path("/repo"))}
+
+        self.assertIn("Claude Code Global", sources)
+        self.assertTrue(sources["Claude Code Global"].endswith("/.claude.json"))
+        self.assertIn("Codex", sources)
+        self.assertTrue(sources["Codex"].endswith("/.codex/config.toml"))
+
     def test_activation_only_manifest_signal_downgrades_to_allow(self) -> None:
         risk = {
             "decision": "warn",
@@ -49,13 +85,29 @@ def test_review_inventory_flag_keeps_config_in_review_tier(self) -> None:
             "activation_score": 0.97,
             "static_score": 0.0,
         }
-        flags = [{"id": "remote-http", "level": "review", "reason": "server connects remotely"}]
+        flags = [{"id": "browser-access", "level": "review", "reason": "server controls a browser"}]
 
         decision, reasons = product_decision_for_config_scan(risk, flags)
 
         self.assertEqual("warn", decision)
         self.assertIn("review-worthy", reasons[0])
 
+    def test_remote_and_env_inventory_do_not_warn_by_default(self) -> None:
+        risk = {
+            "decision": "warn",
+            "activation_score": 0.97,
+            "static_score": 0.0,
+        }
+        flags = [
+            {"id": "remote-http", "level": "info", "reason": "server connects remotely"},
+            {"id": "env-secrets", "level": "info", "reason": "server uses environment variables"},
+        ]
+
+        decision, reasons = product_decision_for_config_scan(risk, flags)
+
+        self.assertEqual("allow", decision)
+        self.assertIn("activation-only", reasons[0])
+
     def test_env_and_remote_server_inventory_flags_do_not_emit_raw_config(self) -> None:
         with tempfile.TemporaryDirectory() as tmp:
             path = Path(tmp) / "mcp.json"
@@ -107,7 +159,7 @@ def test_env_and_remote_server_inventory_flags_do_not_emit_raw_config(self) -> N
 
         self.assertIn("env-secrets", {flag["id"] for flag in flags})
         self.assertIn("remote-http", {flag["id"] for flag in flags})
-        self.assertEqual("warn", payload["results"][0]["decision"])
+        self.assertEqual("allow", payload["results"][0]["decision"])
         self.assertNotIn("secret-value", json.dumps(payload, sort_keys=True))
         self.assertNotIn("redacted_config", payload["results"][0])