FastAPI Safeguard

FastAPI Safeguard scans your routes at startup and fails on security issues — missing auth, exposed sensitive fields, CORS misconfigs, debug mode left on — before your application serves a single request.

Add it once. New unsecured routes fail the build.

Requires Python 3.9+

Highlights

• Baseline lock file — Like package-lock.json, but for security findings. Accept existing tech debt; fail only on new regressions. Every change to security posture shows up as a diff in security_baseline.json.
• Fail fast — Startup exits non-zero if new issues are found. Works as a CI gate out of the box.
• CLI & test helper — fastapi-safeguard check app.main:app scans without running your app; assert_safeguard(app) does the same inside pytest.
• Explicit intent — Mark public endpoints with @open_route. Everything else is expected to have auth.
• Zero runtime overhead — Checks run once at startup, never per-request.
• Extensible — 9 core checks, 8 optional, and a simple base class for your own rules.

Quick Start

pip install fastapi-safeguard

from fastapi import FastAPI, Depends
from fastapi.security import OAuth2PasswordBearer
from fastapi_safeguard import FastAPISafeguard, open_route

app = FastAPI(lifespan=FastAPISafeguard.recommended().lifespan())
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")

@open_route  # Explicitly marked as public — no warning
@app.get("/health")
async def health():
    return {"status": "healthy"}

@app.get("/legacy-endpoint")  # Forgot to add auth — will be flagged!
async def legacy():
    return {"data": "visible to all"}

@app.get("/protected")
async def protected(token: str = Depends(oauth2_scheme)):
    return {"data": "secret"}

First run — accept current state and create a baseline:

SECURITY_BASELINE_UPDATE=1 uvicorn main:app --reload

This creates security_baseline.json:

{
  "schema_version": 1,
  "generated_at": "2025-01-15T10:30:00Z",
  "accepted_findings": [
    "GET /legacy-endpoint has no accepted security dependency"
  ],
  "checks_count": 9
}

Subsequent runs fail only on new findings:

uvicorn main:app --reload

✅ Startup succeeds if findings match baseline ❌ Startup fails if new security issues are detected

---

Table of Contents

• Highlights
• Quick Start
• Why
• Configuration
• Baseline Workflow
• CLI
• Using in Tests
• Decorators
• Security Checks Reference
• OWASP API Top 10 Coverage
• Writing a Custom Check
• CI/CD Integration
• Startup Output Examples
• Environment Variables
• FAQ
• Roadmap
• Contributing
• License

---

Why

FastAPI treats every new endpoint as open by default. That's fine for prototyping. In production it means a forgotten Depends() silently exposes a route — and nobody notices until a pentest or an incident.

The failure modes are mundane and common:

• A mutation endpoint ships without auth.
• A response model is missing, so the ORM object serializes more fields than intended.
• A list endpoint has no pagination, and a single request dumps the whole table.
• CORS is set to * with allow_credentials=True.
• Debug mode is on in a staging environment that shares production data.

FastAPI Safeguard catches these at startup. It enforces that openness is intentional: unsecured routes must either carry a recognized auth dependency, be decorated with @open_route, or be explicitly accepted in the security baseline.

The baseline is the key part. Once you accept a finding, it's pinned. Future runs fail only on new findings. Adding an unsecured route shows up as a diff in the baseline file. Fixing one shows up as a removal. Security posture becomes reviewable in pull requests.

---

Configuration

Custom auth dependencies

Built-in schemes (OAuth2*, HTTPBasic, HTTPBearer, APIKey*) are recognized even when nested inside your own dependencies — the common Depends(get_current_user) pattern, where get_current_user itself depends on oauth2_scheme, works out of the box. Only fully custom auth needs registering so the check recognizes it:

from fastapi_safeguard import (
    FastAPISafeguard,
    DependencySecurityCheck,
    ResponseModelSecurityCheck,
    PaginationEnforcementCheck,
)

safeguard = FastAPISafeguard(checks=[
    DependencySecurityCheck(
        allowed_unsecured=["/openapi.json", "/docs", "/health"],
        extra_dependencies=[my_custom_auth_dep],
    ),
    ResponseModelSecurityCheck(),
    PaginationEnforcementCheck(),
])
app = FastAPI(lifespan=safeguard.lifespan())

Adding optional checks

The recommended() preset includes only the 9 core checks. Add optional ones explicitly:

from fastapi_safeguard import FastAPISafeguard, HTTPSRedirectMiddlewareCheck

safeguard = FastAPISafeguard(checks=[
    *FastAPISafeguard.recommended().checks,
    HTTPSRedirectMiddlewareCheck(),
])
app = FastAPI(lifespan=safeguard.lifespan())

Existing lifespan

If your app already has a lifespan, call run_checks() directly:

from contextlib import asynccontextmanager
from fastapi import FastAPI
from fastapi_safeguard import FastAPISafeguard

safeguard = FastAPISafeguard.recommended()

@asynccontextmanager
async def lifespan(app: FastAPI):
    safeguard.run_checks(app)
    db = await setup_database()
    yield
    await db.close()

app = FastAPI(lifespan=lifespan)

---

Baseline Workflow

Mental model

You do this	The plugin does this
Run app with no baseline	Scans, prints findings, exits with error
Set SECURITY_BASELINE_UPDATE=1	Writes a snapshot of current findings
Add a new unsecured route	Fails: finding not in baseline
Fix a finding	Old entry stays in file until you refresh
Refresh with update flag	Prunes resolved entries, accepts new ones

Typical flow

# 1. First scan (will fail if there are findings)
uvicorn main:app --reload

# 2. Accept the current snapshot
SECURITY_BASELINE_UPDATE=1 uvicorn main:app --reload

# 3. Subsequent runs fail only on NEW findings
uvicorn main:app --reload

# 4. After fixing issues, prune resolved ones
SECURITY_BASELINE_UPDATE=1 uvicorn main:app --reload

You can also set update_baseline=True in code: FastAPISafeguard(..., update_baseline=True).

Reviewing baseline diffs

• Added lines = new risks being accepted.
• Removed lines = fixed issues.
• No change = security posture unchanged.

If a diff adds "GET /internal-metrics has no accepted security dependency", ask: was that endpoint meant to be open? If not, fix it instead of accepting.

Commit the baseline to git. Treat it like a contract — change it only when you're consciously accepting or removing risk.

---

CLI

Scan an app without wiring the lifespan or starting a server — the app is only imported, never run. Use it when you want the security gate in CI (or locally) instead of, or in addition to, the startup gate:

fastapi-safeguard check app.main:app                  # attribute defaults to :app
fastapi-safeguard check app.main:create_app --factory # app factories
fastapi-safeguard check main:app --update-baseline    # accept current findings

Flag	Purpose
--app-dir DIR	Directory to add to the import path (default: .)
--baseline PATH	Baseline file (default: SECURITY_BASELINE_PATH or security_baseline.json)
--update-baseline	Accept current findings and write/refresh the baseline
--factory	Call the imported attribute to get the app

Exit codes: 0 pass, 1 new findings, 2 app could not be loaded. Also available as python -m fastapi_safeguard.

---

Using in Tests

assert_safeguard(app) runs the same scan as a plain assertion — no SystemExit, no stdout noise, baseline respected but never written:

from fastapi_safeguard import assert_safeguard
from app.main import app

def test_security_posture():
    assert_safeguard(app)  # raises AssertionError listing any new findings

It returns a ScanResult for further assertions, and accepts checks=[...], baseline_path=..., or a preconfigured safeguard= for custom setups.

---

Decorators

Decorator	Effect	Other checks still run?	Use case
@open_route	Skips auth check only	Yes	Public status pages, health checks
@disable_security_checks	Skips all checks	No	Internal webhooks, exotic routes

from fastapi_safeguard import open_route, disable_security_checks

@open_route
@app.get("/public-catalog")
async def catalog():
    return {"items": []}

@disable_security_checks
@app.post("/internal-webhook")
async def webhook():
    return {"ok": True}

Decorator placement above or below @app.get(...) doesn't matter — both attach to the same function.

Other ways to accept findings

Strategy	When to use
@open_route	Intentionally public endpoint. Other checks still apply.
@disable_security_checks	Exceptional route. Use sparingly.
Baseline accept	Known tech debt. Persists until pruned.
Omit check from config	The check isn't relevant to your org.

---

Security Checks Reference

Core checks (in recommended())

Check	OWASP	What it catches	Example finding
DependencySecurityCheck	API2, API5	Routes with no auth dependency	GET /items has no accepted security dependency
UnsecuredAllowedMethodsCheck	API5	Unsafe methods on declared-open paths	POST /status exposes unsafe method(s) without security
ResponseModelSecurityCheck	API3	Mutation endpoints missing response_model	POST /users missing response_model for unsafe method(s)
BodyModelEnforcementCheck	API3, API6	Raw dict/list body params (mass assignment risk)	PATCH /users uses non-model raw body param(s): payload
SensitiveFieldExposureCheck	API3	Response fields named password, token, secret, etc.	GET /auth/me response_model exposes potentially sensitive fields: password
SensitiveQueryParamCheck	API3	Sensitive data in query strings	GET /login exposes potentially sensitive data via query params: token
PaginationEnforcementCheck	API4	List endpoints without limit/skip params	GET /reports returns a collection without pagination params
CORSMisconfigurationCheck	API8	Wildcard origins + allow_credentials=True	CORS misconfiguration: allow_origins='*', credentials allowed
DebugModeCheck	API8	app.debug = True	Application running in debug mode

Optional checks

These have higher false positive rates or are typically handled at the infrastructure level:

Check	OWASP	What it catches	Why not default
HTTPSRedirectMiddlewareCheck	API8	Missing HTTPS redirect middleware	Usually handled by load balancer
TrustedHostMiddlewareCheck	API8	Missing host header protection	Usually handled upstream
RateLimitingPresenceCheck	API4	No rate-limiting middleware detected	Often external (API gateway); weak heuristic
ReturnTypeAnnotationCheck	API3	Missing return type annotations	Noisy; not a security issue
WildcardPathCheck	API3, API5	Broad {path:path} params	Legitimate use cases exist
DangerousMethodExposureCheck	—	TRACE/CONNECT methods	Rarely exposed via FastAPI
SSRFParameterCheck	—	Params named url, uri, target, etc.	Too many false positives
AdminRouteOpenCheck	—	Unprotected /admin* routes	Heuristic based on path naming

---

OWASP API Top 10 Coverage

OWASP	Risk	Core checks	Optional checks
API2	Broken Authentication	DependencySecurityCheck	—
API3	Excessive Data Exposure	ResponseModelSecurityCheck, BodyModelEnforcementCheck, SensitiveFieldExposureCheck, SensitiveQueryParamCheck	ReturnTypeAnnotationCheck, WildcardPathCheck
API4	Unrestricted Resource Consumption	PaginationEnforcementCheck	RateLimitingPresenceCheck
API5	Broken Function Level Authorization	DependencySecurityCheck, UnsecuredAllowedMethodsCheck	WildcardPathCheck
API6	Mass Assignment	BodyModelEnforcementCheck	—
API8	Security Misconfiguration	CORSMisconfigurationCheck, DebugModeCheck	HTTPSRedirectMiddlewareCheck, TrustedHostMiddlewareCheck

---

Writing a Custom Check

from typing import Optional

from fastapi.routing import APIRoute
from fastapi_safeguard import FastAPISafeguard, RouteCheck

class MaxPathDepthCheck(RouteCheck):
    """Flag routes with deeply nested paths (more than 4 segments)."""
    CATEGORY = "routing"
    OWASP = ["API5"]

    def _analyze(self, route: APIRoute) -> Optional[str]:
        depth = len(route.path.strip("/").split("/"))
        if depth > 4:
            return f"{route.path} has {depth} path segments (max 4)"
        return None

safeguard = FastAPISafeguard(checks=[
    *FastAPISafeguard.recommended().checks,
    MaxPathDepthCheck(),
])

Rules:

• Subclass RouteCheck (not SecurityCheck) and implement _analyze — the base class handles @disable_security_checks and allowed_unsecured paths for you.
• Return None when compliant, a string when not.
• Finding strings must be stable and deterministic — the baseline matches on exact text.
• Set CATEGORY and OWASP for the summary output.

---

CI/CD Integration

# GitHub Actions — scans the app without running it; exits 1 on new findings
- name: Security route check
  run: |
    pip install fastapi-safeguard
    fastapi-safeguard check main:app

To accept a new baseline in a controlled PR:

- name: Update baseline (manual trigger)
  if: github.event_name == 'workflow_dispatch'
  run: |
    fastapi-safeguard check main:app --update-baseline
    git add security_baseline.json
    git commit -m "chore: update security baseline" || echo "No changes"

---

Startup Output Examples

New findings (startup fails):

Category Summary:
Category            Total   New  Accepted  OWASP
auth                   1     1         0   API2/API5
schema                 2     1         1   API3/API6

❌ Security check failed: new findings detected (not in baseline):
  + POST /users missing response_model for unsafe method(s)
  + GET /items has no accepted security dependency

To accept current findings run with SECURITY_BASELINE_UPDATE=1 or set update_baseline=True.

All findings accepted:

Category Summary:
Category            Total   New  Accepted  OWASP
auth                   1     0         1   API2/API5
schema                 2     0         2   API3/API6

✅ All security findings match accepted baseline (3 accepted).

Clean:

✅ All security checks passed (0 findings, 12 routes, 14 checks).

---

Environment Variables

Variable	Purpose	Default
SECURITY_BASELINE_PATH	Path to baseline file	security_baseline.json
SECURITY_BASELINE_UPDATE	Set to 1 to accept current findings and write/refresh the baseline	0

---

FAQ

Does this replace runtime auth? No. It catches common omissions at startup. You still need runtime authorization.

Will it block production if I forget to update the baseline? Yes, intentionally. Incorporate baseline updates into the PR that introduces the change.

Can I have different baselines per environment? Yes — set SECURITY_BASELINE_PATH to point at different files.

What if I rename a route? The old finding becomes resolved; refresh with SECURITY_BASELINE_UPDATE=1 to prune it.

How do I disable a specific check? Omit it from the checks=[...] list when constructing FastAPISafeguard.

---

Roadmap

• Severity levels (warn vs error) and downgrade mechanism
• JSON output for machine parsing (SARIF-compatible)
• Category include/exclude filters via env vars
• @warn_only decorator for non-fatal findings on selected routes
• Scope / role enforcement scaffold

---

Contributing

1. Fork & branch (feat/my-improvement).
2. Keep PRs focused — one conceptual change per PR.
3. Update README examples if behavior changes.
4. Prefer small, composable checks over multi-purpose ones.
5. Finding strings must be deterministic — no timestamps or randomness.

---

License

MIT — see LICENSE.