403-bypass

Developed by JOJIN JOHN

403-bypass is a command-line tool for testing HTTP access-control bypasses and parser inconsistencies around 401, 403, and related responses.

The tool is designed for practical web security work: bug bounty, penetration testing, security reviews, and regression testing of access-control rules. It automates a broad set of request mutations, captures a baseline, filters common false positives, and highlights the responses most likely to represent a meaningful bypass.

What It Does

Given a target URL, nomore403:

1. Sends a baseline request to capture the blocked response.
2. Optionally auto-calibrates against non-existent paths to learn the target's default error behavior.
3. Runs a set of bypass techniques that mutate the request path, method, headers, or wire format.
4. Scores and groups the results to reduce noise.
5. Emits replayable evidence, including curl commands for interesting findings.

This tool does not "break authentication" by itself. It helps find differences between how frontends, proxies, WAFs, CDNs, application routers, and backends interpret the same request.

Features

• Baseline-driven comparison against the blocked response
• Auto-calibration to reduce false positives from default 404 or parent-path responses
• Scored output with separate summaries for likely bypasses and interesting variations
• False-positive filtering using status, length, body hash, HTML title, body class, and WAF/block-page detection
• Replay and reproducibility for high-value findings
• Retry and backoff for transient network failures
• Concurrent execution with per-technique progress
• Raw HTTP support for request forms that net/http normalizes away
• JSON and JSONL output for pipelines and post-processing
• Input from a single URL, URL files, stdin, or request files

Installation

Build from source

git clone https://github.com/jojin1709/403-bypass.git
cd 403-bypass
go build -o nomore403

Build on Windows PowerShell

git clone https://github.com/jojin1709/403-bypass.git
cd 403-bypass
go build -o nomore403.exe
.\nomore403.exe --help

If Go cannot write to the default cache folder on Windows, use local cache folders before testing or building:

$env:GOCACHE="$PWD\.gocache"
$env:GOPATH="$PWD\.gopath"
go test ./...
go build -o nomore403.exe

The payloads/ directory is read at runtime. Keep it beside the executable, or point the tool to it with -f.

Requirements

• Go 1.24 or later to build from source
• curl available in PATH for techniques that depend on it, such as:
  • http-versions
  • http-parser
  • absolute-uri

Most techniques work without curl.

Quick Start

Basic scan:

./nomore403 -u https://target.tld/admin

Use a proxy and verbose output:

./nomore403 -u https://target.tld/admin -x http://127.0.0.1:8080 -v

Run only selected techniques:

./nomore403 -u https://target.tld/admin -k headers,absolute-uri,raw-desync

Read targets from stdin:

cat urls.txt | ./nomore403

Use a Burp-style request file:

./nomore403 --request-file request.txt

Write machine-readable output:

./nomore403 -u https://target.tld/admin --jsonl -o findings.jsonl

Windows PowerShell examples

Run the executable from the current folder:

.\nomore403.exe --help

Run a low-noise scan:

.\nomore403.exe -u https://target.tld/admin -k headers,endpaths,midpaths --top 10

Run slower against WAF-protected authorized targets:

.\nomore403.exe -u https://target.tld/admin -k headers,endpaths --max-goroutines 2 --delay 1000 --top 10

Save JSONL output:

.\nomore403.exe -u https://target.tld/admin --jsonl -o findings.jsonl

Use a Burp-style raw request file:

.\nomore403.exe --request-file .\request.txt --jsonl -o findings.jsonl

Example Output

target: https://target.tld/admin   method: GET   frontend: AWS ELB/ALB   payloads: payloads

calib: 404 | 1245b | +/-50 | frag 703b

BASELINE
  default       403    520 bytes    https://target.tld/admin

FINDINGS
  hdr-ip     100! 200   2048 bytes   X-Original-URL: /
  abs-uri     26. 403    236 bytes   request-target: https://target.tld/admin
  http        18. 400    122 bytes   HTTP/2

no visible results: 17 techniques

============== LIKELY BYPASS ==============
[!100 HIGH] Header injection (IP) 403=>200    2048b
         why: status 403->200, len delta 1528, body changed, type changed
        item: X-Original-URL: /
        curl: curl -i -sS -k -H 'User-Agent: nomore403' -H 'X-Original-URL: /' 'https://target.tld/admin'

How to Read the Output

Main technique output

Each visible line is a response that differed enough from the baseline to survive filtering.

Typical fields:

• technique alias, for example hdr-ip, abs-uri, or parser
• compact score marker such as 18., 61+, or 100!
• final response status
• response size
• item or payload used

Summary transitions

The final summaries show baseline-to-result transitions:

• 403=>200 usually deserves immediate attention
• 403=>302 can be interesting, but may still resolve back into an auth barrier
• 403->400 or 403->404 usually indicate parser or routing differences rather than a bypass

Summaries

At the end of the run, nomore403 prints:

• LIKELY BYPASS
  • highest-scoring results
  • includes reproducible curl
• INTERESTING VARIATIONS
  • meaningful parser or routing differences that are worth manual review
• no visible results
  • count of techniques that ran but produced no visible output after filtering

Scoring Model

Scoring is heuristic. It is intended to prioritize results, not to prove exploitation.

The tool generally rewards:

• transitions to 2xx
• transitions to 3xx
• large body-length changes
• body hash changes
• HTML title changes
• body-class changes, such as waf-block to html or login to json
• Location changes
• anomalous redirects that do not appear to resolve into a login or denied flow
• differences that survive replay

The tool generally down-ranks:

• near-identical responses
• 200 responses that have the same body hash as the blocked baseline
• responses with the same HTML title and same body class as the blocked baseline
• Cloudflare, WAF, or generic block pages
• repeated parser noise
• unstable replay results
• empty-body redirects that appear to lead back into access control
• many 400 and 404 cases unless the response also changes substantially

Recommended interpretation:

• HIGH: likely actionable; review first
• MED: plausible candidate; usually worth manual replay
• LOW: parser difference, routing anomaly, or lower-confidence behavior

False-positive filtering

Many sites return 200 for a rewritten or fragment-stripped URL while still serving the same error, homepage, login page, or WAF block page. The tool now fingerprints responses with:

• body hash
• HTML <title>
• body class, for example waf-block, login, access-denied, not-found, json, or html
• WAF/block-page signals from body text and headers such as CF-Ray

This means a result like 403->200 is no longer treated as strong by status alone. If it has the same body, same title, or the same WAF/block page as the baseline, it is pushed down into low-confidence output.

In JSON and JSONL output, check these fields:

{
  "status_code": 200,
  "score": 18,
  "likelihood": "low",
  "score_reason": "status 403->200, same body as baseline, same body class",
  "body_hash": "9f31...",
  "title": "Oooops! WAF has blocked the action.",
  "body_class": "waf-block",
  "waf_block": true
}

Treat waf_block: true, same body as baseline, or same body class as signs that the result probably needs manual review before calling it a bypass.

Auto-Calibration

Auto-calibration is enabled by default in non-verbose mode.

It sends requests to several non-existent paths and builds a baseline for the target's default error behavior. It also performs a fragment-based calibration request to reduce false positives caused by fragment-stripped paths.

Use these flags to control it:

• --no-calibrate
  • compare only against the default blocked response
• --strict-calibrate
  • also compare body hash and key headers such as Location, Content-Type, and Server

Techniques

The tool runs all techniques by default unless you specify -k.

Method and verb mutations

• verbs
  • alternative HTTP methods from payloads/httpmethods
• verbs-case
  • randomized casing of HTTP methods
• method-override
  • query, header, and body-based method override patterns

Header- and trust-based mutations

• headers
  • umbrella technique covering:
    • IP trust headers
    • simple headers
    • Host variations
• hop-by-hop
  • hop-by-hop stripping tricks using Connection
• header-confusion
  • rewrite and path-override headers such as X-Original-URL
• host-override
  • host override and forwarded-host variants
• forwarded-trust
  • Forwarded, Client-IP, Cluster-Client-IP, and related trust chains
• proto-confusion
  • X-Forwarded-Proto, X-Forwarded-Port, and related scheme hints
• ip-encoding
  • localhost and trusted-address variants in dotted, integer, hex, and IPv6 forms

Path and normalization mutations

• endpaths
  • suffix and end-of-path mutations from payloads/endpaths
• midpaths
  • path insertion and traversal-style mutations from payloads/midpaths
• double-encoding
  • encoded path variants, including aggressive double-encoding forms
• unicode
  • %uXXXX and overlong UTF-8 path variants
• path-case
  • path segment case switching
• path-normalization
  • dot-segment and semicolon normalization variants
• suffix-tricks
  • suffix and extension tricks such as .json, .css, ;index.html, and format-style query toggles
• payload-position
  • inserts payloads at explicitly marked positions in the URL

Frontend and wire-format mutations

• http-versions
  • compares the same request across HTTP/1.0 and HTTP/2
• http-parser
  • sends a deliberately minimal curl request to expose client/frontend parser differences separately from http-versions
• absolute-uri
  • uses absolute-form request targets through curl --request-target
• raw-duplicates
  • duplicate security-relevant headers with raw HTTP
• raw-authority
  • duplicate or conflicting authority and host signals
• raw-desync
  • request forms aimed at frontend/backend parsing differences, including conflicting transfer semantics

Raw HTTP Behavior

Some techniques need wire-level control that Go's net/http client does not provide. Those techniques use the raw HTTP engine.

Raw techniques currently include:

• raw-duplicates
• raw-authority
• raw-desync
• some %uXXXX unicode path requests

Notes:

• raw requests are sent automatically where needed
• raw techniques do not currently support upstream proxies
• raw behavior is useful for:
  • duplicate headers
  • exact request targets
  • transfer-encoding and content-length edge cases

Fingerprinting and Technique Order

The tool may infer frontend hints such as:

• AWS ELB / ALB
• CloudFront
• Cloudflare
• Nginx
• Envoy
• Apache
• IIS

These hints are used to improve technique ordering and output context.

Important:

• fingerprinting does not disable techniques by default
• unless you use -k, the tool still runs the full default technique set

Reproducibility and Replay

High-value results are replayed automatically in the final summary.

The replay output helps answer:

• did the behavior repeat?
• did it keep the same status and response shape?
• is this likely stable enough to investigate or report?

The final summary includes:

• replay counts such as 1/1 or 2/2 matched on replay
• a replayable curl command for interesting results

Input Modes

Single target

./nomore403 -u https://target.tld/admin

File containing URLs

./nomore403 -u targets.txt

Stdin

cat targets.txt | ./nomore403

Request file

./nomore403 --request-file request.txt

Request files are useful when replaying traffic captured in Burp or another proxy.

Custom Payload Positions

Use --payload-position when you want to inject payloads at explicit markers instead of relying only on built-in endpath and midpath mutation logic.

Example:

./nomore403 -u 'https://target.tld/§100§/admin/§200§' -p §

Common Workflows

Fast triage

./nomore403 -u https://target.tld/admin

Investigate with an intercepting proxy

./nomore403 -u https://target.tld/admin -x http://127.0.0.1:8080 -v

Run only raw and frontend-focused techniques

./nomore403 -u https://target.tld/admin -k absolute-uri,raw-duplicates,raw-authority,raw-desync

Test a custom trusted IP

./nomore403 -u https://target.tld/admin -i 10.0.0.5

Save JSONL for later analysis

./nomore403 -u https://target.tld/admin --jsonl -o findings.jsonl

Windows PowerShell with JSON output

.\nomore403.exe -u https://target.tld/admin --json -o findings.json

Linux/macOS request-file mode

./nomore403 --request-file request.txt -k headers,endpaths --jsonl -o findings.jsonl

Windows request-file mode

.\nomore403.exe --request-file .\request.txt -k headers,endpaths --jsonl -o findings.jsonl

Build and test locally

Linux/macOS:

go test ./...
go build -o nomore403

Windows PowerShell:

$env:GOCACHE="$PWD\.gocache"
$env:GOPATH="$PWD\.gopath"
go test ./...
go build -o nomore403.exe

GitHub Actions

The repository includes:

• .github/workflows/test.yml
  • runs go test ./...
  • runs go build ./...
  • triggers on pushes to main and pull requests
• .github/workflows/release.yml
  • builds Windows .exe
  • builds Linux amd64
  • builds macOS amd64 and arm64
  • packages the payloads/ folder with each binary
  • publishes artifacts automatically when you push a version tag

Create a release from your terminal:

git tag v0.1.0
git push origin v0.1.0

On Windows PowerShell:

git tag v0.1.0
git push origin v0.1.0

Flags

Run ./nomore403 --help for the full up-to-date CLI.

Key flags:

• -u, --uri
  • target URL, file of URLs, or input path
• -k, --technique
  • comma-separated list of techniques to run
• -x, --proxy
  • upstream proxy
• -H, --header
  • add custom headers
• -i, --bypass-ip
  • IP or hostname used in trust-header techniques
• -v, --verbose
  • show all visible results, not just filtered output
• --json
  • write structured JSON
• --jsonl
  • write one JSON object per result
• --no-calibrate
  • disable auto-calibration
• --strict-calibrate
  • compare more response fields during filtering
• --retry-count
  • number of retries for transient errors
• --retry-backoff-ms
  • exponential backoff base in milliseconds
• --host-delay
  • delay between batched targets on the same host
• --top-score-min
  • minimum score for LIKELY BYPASS
• --variation-score-min
  • minimum score for INTERESTING VARIATIONS
• --top
  • maximum number of entries per summary section, or 0 to disable summaries

Output Formats

Human-readable terminal output

Default mode is optimized for interactive review and triage.

JSON

Use --json for a single structured document.

JSON Lines

Use --jsonl when you want to:

• process results incrementally
• store evidence in pipelines
• import findings into your own tooling

Payload Files

The payloads/ directory contains lists used by several techniques.

Current files include:

• httpmethods
• headers
• ips
• simpleheaders
• endpaths
• midpaths
• useragents

You can customize these files to fit your targets or workflow.

Limitations

• raw HTTP techniques do not currently support upstream proxies
• scoring is heuristic and can produce false positives or false negatives
• redirect scoring currently uses heuristics on the immediate redirect response, not a fully followed redirect chain
• some techniques depend on target-specific behavior and may appear noisy on heavily normalized stacks
• curl-based techniques require curl in PATH

Security and Responsible Use

Use this tool only on systems you are authorized to test.

The authors and contributors are not responsible for misuse. You are responsible for complying with applicable law, program rules, and organizational policy.

Contributing

Contributions are welcome.

Useful contribution areas include:

• bug fixes
• better payloads
• new bypass techniques
• raw HTTP improvements
• frontend fingerprinting
• documentation and examples

Before contributing a technique, prefer:

• a clearly distinct parsing or trust-boundary behavior
• reproducible evidence
• tests that verify the request shape or replay behavior

License

This project is released under the MIT License. See LICENSE.