easyapi-django 1.2.2__tar.gz → 1.3.0__tar.gz

{easyapi_django-1.2.2/easyapi_django.egg-info → easyapi_django-1.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: easyapi-django
-Version: 1.2.2
+Version: 1.3.0
 Summary: Zero-effort MCP server (and REST API) generator from Django models.
 Author-email: Stamatios Stamou Jr <bushier.outsets.0c@icloud.com>
 License-Expression: MIT
@@ -235,7 +235,8 @@ Defaults are demo-friendly. Production deployments should:
 - **Tune session TTLs.** `SESSION_TTL` (cookies, default 1800s) and `API_SESSION_TTL` (api-key cache, default 300s) both slide on use. Pick numbers that match your security/UX trade-off.
 - **Never ship `DEBUG = True`.** Under `DEBUG`, edge security (scanner blocking, 4xx flood detection), rate limiting, credential-path defense and abuse blocking are all bypassed for dev ergonomics. The framework logs a loud startup warning when the bypass is active. To decouple from `DEBUG` — e.g. a staging box mirroring production — set `MCP = {'SECURITY_BYPASS': False}` to force protections on (or `True` to bypass without `DEBUG`).
 - **Trust your proxies explicitly.** IP-based rate limits and abuse blocking key on the client IP. `X-Forwarded-For` / `X-Real-IP` are only honored when `REMOTE_ADDR` is in `TRUSTED_PROXIES = ['10.0.0.0/8', ...]`; otherwise `REMOTE_ADDR` wins. Leave it unset behind a proxy and every request looks like it comes from the proxy.
-- **Bound the metrics surface.** The `/metrics` endpoint queries any model in the registry by default. Set `METRICS_MODELS = ['app_Model', ...]` to allowlist which models are queryable. References to credential-named columns (in filters, groups, order or calc) are always rejected.
+- **Bound the metrics surface.** The `/metrics` endpoint queries any model in the registry by default. Set `METRICS_MODELS = ['app_Model', ...]` to allowlist which models are queryable. References to credential-named columns (in filters, groups, order or calc) are always rejected — the per-resource escape hatch does not reach `/metrics`.
+- **Keep the escape hatch narrow.** Columns whose names look like credentials (`password`, `api_key`, `token`, `otp`, `secret`, …) are masked as `'*********'` on every response, on top of whatever `sensitive_fields` lists. When the heuristic catches a false positive, opt it out per surface: `edit_allow_sensitive_fields = ['token_count']` returns it verbatim on `GET /resource/<id>/` while LIST still masks it. `list_allow_sensitive_fields` does the same for LIST, where every row carries the value — reach for the detail-only form first.
 - **OAuth clients are public by default.** MCP clients (Claude.ai, ChatGPT, ...) authenticate with PKCE — list them under `MCP['OAUTH_CLIENTS']` with no `client_secret`. A client becomes confidential only with an explicit secret; `/register` never hands out a confidential secret (provision those out of band).
 
 ---

{easyapi_django-1.2.2 → easyapi_django-1.3.0}/README.md

@@ -191,7 +191,8 @@ Defaults are demo-friendly. Production deployments should:
 - **Tune session TTLs.** `SESSION_TTL` (cookies, default 1800s) and `API_SESSION_TTL` (api-key cache, default 300s) both slide on use. Pick numbers that match your security/UX trade-off.
 - **Never ship `DEBUG = True`.** Under `DEBUG`, edge security (scanner blocking, 4xx flood detection), rate limiting, credential-path defense and abuse blocking are all bypassed for dev ergonomics. The framework logs a loud startup warning when the bypass is active. To decouple from `DEBUG` — e.g. a staging box mirroring production — set `MCP = {'SECURITY_BYPASS': False}` to force protections on (or `True` to bypass without `DEBUG`).
 - **Trust your proxies explicitly.** IP-based rate limits and abuse blocking key on the client IP. `X-Forwarded-For` / `X-Real-IP` are only honored when `REMOTE_ADDR` is in `TRUSTED_PROXIES = ['10.0.0.0/8', ...]`; otherwise `REMOTE_ADDR` wins. Leave it unset behind a proxy and every request looks like it comes from the proxy.
-- **Bound the metrics surface.** The `/metrics` endpoint queries any model in the registry by default. Set `METRICS_MODELS = ['app_Model', ...]` to allowlist which models are queryable. References to credential-named columns (in filters, groups, order or calc) are always rejected.
+- **Bound the metrics surface.** The `/metrics` endpoint queries any model in the registry by default. Set `METRICS_MODELS = ['app_Model', ...]` to allowlist which models are queryable. References to credential-named columns (in filters, groups, order or calc) are always rejected — the per-resource escape hatch does not reach `/metrics`.
+- **Keep the escape hatch narrow.** Columns whose names look like credentials (`password`, `api_key`, `token`, `otp`, `secret`, …) are masked as `'*********'` on every response, on top of whatever `sensitive_fields` lists. When the heuristic catches a false positive, opt it out per surface: `edit_allow_sensitive_fields = ['token_count']` returns it verbatim on `GET /resource/<id>/` while LIST still masks it. `list_allow_sensitive_fields` does the same for LIST, where every row carries the value — reach for the detail-only form first.
 - **OAuth clients are public by default.** MCP clients (Claude.ai, ChatGPT, ...) authenticate with PKCE — list them under `MCP['OAUTH_CLIENTS']` with no `client_secret`. A client becomes confidential only with an explicit secret; `/register` never hands out a confidential secret (provision those out of band).
 
 ---

{easyapi_django-1.2.2 → easyapi_django-1.3.0}/easyapi/base.py

@@ -40,7 +40,7 @@ from .helpers import (
     search_regex,
 )
 from .schemas import is_pydantic_model, serialize_output, validate_input
-from .sensitive import looks_sensitive
+from .sensitive import is_sensitive
 from .serializer import CustomJSONEncoder
 from .util import validate_session_key
 from .rate_limit import RateLimiter
@@ -186,13 +186,21 @@ class BaseResource(CoreResource):
     description = None
 
     # Field names that should never leave the server. Listed names are
-    # appended to ``list_exclude_fields`` and ``edit_exclude_fields`` at
-    # init time. The framework already filters a baseline of obviously
-    # sensitive names (``password``, ``api_key``, ``token``, ``otp``,
-    # ``secret``, …) — use this for project-specific columns (audit
-    # fields, internal flags, server-assigned IDs).
+    # masked as ``*********`` on serialization. The framework already
+    # filters a baseline of obviously sensitive names (``password``,
+    # ``api_key``, ``token``, ``otp``, ``secret``, …) — use this for
+    # project-specific columns (audit fields, internal flags,
+    # server-assigned IDs).
     sensitive_fields = None
 
+    # Escape hatch: names here are serialized verbatim, overriding both
+    # ``sensitive_fields`` and the name heuristic. Per surface, because a
+    # column safe to return on a single-object read is rarely safe to
+    # return on every row of an unbounded list. ``edit_*`` governs the
+    # detail payload (``_get_obj``), not the write path.
+    list_allow_sensitive_fields = None
+    edit_allow_sensitive_fields = None
+
     account_db = 'default'
     cache = False
     # Default TTL: settings.CACHE_TTL when set, else 120s.
@@ -307,6 +315,12 @@ class BaseResource(CoreResource):
         # Normalise once so downstream serialisation can use a set
         # without recomputing per-row.
         self.sensitive_fields = set(self.sensitive_fields) if self.sensitive_fields else set()
+        self.list_allow_sensitive_fields = (
+            set(self.list_allow_sensitive_fields) if self.list_allow_sensitive_fields else set()
+        )
+        self.edit_allow_sensitive_fields = (
+            set(self.edit_allow_sensitive_fields) if self.edit_allow_sensitive_fields else set()
+        )
         self.update_fields = list(self.update_fields) if self.update_fields else []
         self.create_fields = list(self.create_fields) if self.create_fields else []
         self.routes = list(self.routes) if self.routes else []
@@ -1325,7 +1339,7 @@ class BaseResource(CoreResource):
                 if field in self.list_exclude_fields:
                     continue
 
-                if looks_sensitive(field) or field in self.sensitive_fields:
+                if is_sensitive(field, self.sensitive_fields, self.list_allow_sensitive_fields):
                     result[field] = '*********'
                 else:
                     result[field] = getattr(row, field, None)
@@ -1449,7 +1463,7 @@ class BaseResource(CoreResource):
                 result[key].append(item)
 
         for field in self.edit_fields:
-            if looks_sensitive(field) or field in self.sensitive_fields:
+            if is_sensitive(field, self.sensitive_fields, self.edit_allow_sensitive_fields):
                 result[field] = '*********'
             else:
                 result[field] = getattr(self.obj, field, None)

{easyapi_django-1.2.2 → easyapi_django-1.3.0}/easyapi/mcp/protocol.py

@@ -31,7 +31,7 @@ import logging
 
 from pydantic import BaseModel, ValidationError
 
-from ..sensitive import looks_sensitive
+from ..sensitive import is_sensitive
 from .schema import validate_input
 
 logger = logging.getLogger(__name__)
@@ -75,7 +75,7 @@ def _public_tools(tools):
             for t in tools]
 
 
-def _serialize_result(value, sensitive_fields=()):
+def _serialize_result(value, sensitive_fields=(), allow_sensitive_fields=()):
     if isinstance(value, BaseModel):
         payload = value.model_dump()
     elif isinstance(value, list) and value and isinstance(value[0], BaseModel):
@@ -84,20 +84,20 @@ def _serialize_result(value, sensitive_fields=()):
         payload = value
     # Always scrub: the credential-name baseline applies even when the
     # toolset declares no ``sensitive_fields``.
-    _scrub(payload, set(sensitive_fields))
+    _scrub(payload, set(sensitive_fields), set(allow_sensitive_fields))
     return payload
 
 
-def _scrub(obj, blocked):
+def _scrub(obj, blocked, allowed):
     if isinstance(obj, dict):
         for key in list(obj.keys()):
-            if key in blocked or looks_sensitive(key):
+            if is_sensitive(key, blocked, allowed):
                 obj.pop(key)
             else:
-                _scrub(obj[key], blocked)
+                _scrub(obj[key], blocked, allowed)
     elif isinstance(obj, list):
         for item in obj:
-            _scrub(item, blocked)
+            _scrub(item, blocked, allowed)
 
 
 def _caller_scopes(request):
@@ -327,7 +327,8 @@ async def _dispatch_tool_call(params, tools, request):
         logger.exception('mcp.after_call.error tool=%s', tool_name)
 
     sensitive = getattr(toolset_cls, 'sensitive_fields', None) or ()
-    payload = _serialize_result(ret, sensitive_fields=sensitive)
+    allowed = getattr(toolset_cls, 'allow_sensitive_fields', None) or ()
+    payload = _serialize_result(ret, sensitive_fields=sensitive, allow_sensitive_fields=allowed)
     return _tool_response(
         {'tool': tool_name, 'code': 'OK', 'data': payload},
         is_error=False,

{easyapi_django-1.2.2 → easyapi_django-1.3.0}/easyapi/sensitive.py

@@ -35,3 +35,15 @@ def looks_sensitive(column_name):
     if any(name.endswith(s) for s in _SENSITIVE_SUFFIXES):
         return True
     return False
+
+
+def is_sensitive(field, sensitive_fields=(), allow_sensitive_fields=()):
+    """Whether ``field`` must be withheld on the surface the caller serves.
+
+    ``allow_sensitive_fields`` is an escape hatch and wins over both the
+    per-resource deny list and the name heuristic — an explicit opt-in is
+    the only way a credential-named column ever leaves the server.
+    """
+    if field in allow_sensitive_fields:
+        return False
+    return looks_sensitive(field) or field in sensitive_fields

{easyapi_django-1.2.2 → easyapi_django-1.3.0/easyapi_django.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: easyapi-django
-Version: 1.2.2
+Version: 1.3.0
 Summary: Zero-effort MCP server (and REST API) generator from Django models.
 Author-email: Stamatios Stamou Jr <bushier.outsets.0c@icloud.com>
 License-Expression: MIT
@@ -235,7 +235,8 @@ Defaults are demo-friendly. Production deployments should:
 - **Tune session TTLs.** `SESSION_TTL` (cookies, default 1800s) and `API_SESSION_TTL` (api-key cache, default 300s) both slide on use. Pick numbers that match your security/UX trade-off.
 - **Never ship `DEBUG = True`.** Under `DEBUG`, edge security (scanner blocking, 4xx flood detection), rate limiting, credential-path defense and abuse blocking are all bypassed for dev ergonomics. The framework logs a loud startup warning when the bypass is active. To decouple from `DEBUG` — e.g. a staging box mirroring production — set `MCP = {'SECURITY_BYPASS': False}` to force protections on (or `True` to bypass without `DEBUG`).
 - **Trust your proxies explicitly.** IP-based rate limits and abuse blocking key on the client IP. `X-Forwarded-For` / `X-Real-IP` are only honored when `REMOTE_ADDR` is in `TRUSTED_PROXIES = ['10.0.0.0/8', ...]`; otherwise `REMOTE_ADDR` wins. Leave it unset behind a proxy and every request looks like it comes from the proxy.
-- **Bound the metrics surface.** The `/metrics` endpoint queries any model in the registry by default. Set `METRICS_MODELS = ['app_Model', ...]` to allowlist which models are queryable. References to credential-named columns (in filters, groups, order or calc) are always rejected.
+- **Bound the metrics surface.** The `/metrics` endpoint queries any model in the registry by default. Set `METRICS_MODELS = ['app_Model', ...]` to allowlist which models are queryable. References to credential-named columns (in filters, groups, order or calc) are always rejected — the per-resource escape hatch does not reach `/metrics`.
+- **Keep the escape hatch narrow.** Columns whose names look like credentials (`password`, `api_key`, `token`, `otp`, `secret`, …) are masked as `'*********'` on every response, on top of whatever `sensitive_fields` lists. When the heuristic catches a false positive, opt it out per surface: `edit_allow_sensitive_fields = ['token_count']` returns it verbatim on `GET /resource/<id>/` while LIST still masks it. `list_allow_sensitive_fields` does the same for LIST, where every row carries the value — reach for the detail-only form first.
 - **OAuth clients are public by default.** MCP clients (Claude.ai, ChatGPT, ...) authenticate with PKCE — list them under `MCP['OAUTH_CLIENTS']` with no `client_secret`. A client becomes confidential only with an explicit secret; `/register` never hands out a confidential secret (provision those out of band).
 
 ---

{easyapi_django-1.2.2 → easyapi_django-1.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "easyapi-django"
-version = "1.2.2"
+version = "1.3.0"
 authors = [
   { name="Stamatios Stamou Jr", email="bushier.outsets.0c@icloud.com" },
 ]