See every path from outside input (requests, CLI arguments, files) to dangerous operations like shell commands, file writes and network calls, function by function. Free: your attack surface, e.g. “132 dangerous functions, 67 reachable from outside input.” AI security review with Pro.
02 · UnderstandVisual Callgraph Navigation
Click through an interactive map of your code: who calls a function, what it calls, and what a change can break several calls away. Get to know an unfamiliar codebase in minutes, not days.
03 · TestRun the right tests for every commit, in your browser
For each commit, CodeMD finds the tests that actually cover the changed code and runs them for you. No local setup, no hunting for commands. A pass means those lines really ran.
04 · CatchBugs, code search and a commit timeline
Automatic bug detection, fast search across the whole repository, and a day-by-day graph of what changed, so you know where to look first.
Step 1 · Free Repo Analysis
Analyze your repository — choose one of the three ways below.
Connect GitHub and pick a repository below, paste a GitHub URL, or upload a code archive. Any one of the three starts the same free analysis.
In minutes, get a daily commit graph, a navigable callgraph, fast code search, and automatic bug detection — all free.
For every commit, CODEmd.dev also finds and runs the tests that already cover your change — free, no AI needed. It catches what a quick diff read misses, like a break several calls away from the function you actually edited.
Every real test that touches the change — not just a filename match.
Coverage-confirmed: a pass means those lines actually ran.
Catches breaks anywhere down the call chain, not just the edited function.
Option A — Connect GitHub
Sign in to GitHub, choose a repository, and get commit graphs, callgraphs, bug detection, and code search instantly.
Each Analyze opens the analysis at codemd.dev/analyze in a new window, with your repository already loading.
One plugin. Three checks.
Get a clear answer to the questions that matter before you merge.
⌁
Security Call Path Analysis
AI-written code can open a door an attacker may find or expose customer data. CodeMD flags new risky paths in the pull request, so you can fix them while it costs a code review, not a breach.
Looks for commands on your server, SQL injection, eval, XSS, path traversal, open redirects, file writes and outbound requests, which helps lower the chance of an attacker getting in
Flags personal data that outside input can reach, which helps you catch a possible leak before it becomes a breach
Shows the whole path, so review is quicker
▶
Execute Existing Tests
A change can break something several calls from the line you edited. CodeMD runs the tests connected to it, so bugs are more likely to surface before your customers see them.
Can mean less waiting on CI and fewer Actions minutes than the whole suite
Uses your repository's own tests, with no hunting for commands
Reports pass or fail on GitHub, with evidence the changed lines ran
+
Create Missing Tests
Generate focused tests when new behavior is not covered, giving reviewers evidence instead of assumptions.
Each check looks for a way an attacker, or a manipulated AI, could get in. Free checks never send your code to an AI model; the AI review runs only on a paid Pro plan, and only when you or your pull request ask for it.
Call-path scan Free · every pull request
Follows each outside input (a web request, a webhook, a command-line argument) through your functions to the risky things it can reach.
Commands on your server, SQL built from text, eval, HTML written into a page, file reads, writes and deletes, redirects, git operations, outbound requests
Fails the pull request when a change opens a new path nobody reviewed
Personal-data (PII) exposure Free · every pull request
Finds where your code holds passwords, health, financial, government-ID, biometric and contact data, and whether a route that outside input can reach returns it, logs it or puts it in a URL.
Checks for a sign-in guard in front of it
Fails the pull request when a change newly exposes it
Code weakness rules Free · Analyze Repo
Reads every source file for weaknesses that are a single line of code, not a risky call.
Keys, tokens and private keys written into the source, including AI provider keys
Debug mode left on, weak random numbers used for credentials, plain-text or MD5/SHA-1 passwords, unescaped template output, request data copied straight into a record, redirects taken from a request field
AI-attack rules Free · Analyze Repo
Looks for the ways software that uses an AI model gets turned against its owner.
Model output run as code, a shell command or a database query
Model output written into a page without escaping
Request, issue or repository text pasted straight into a prompt, the opening for prompt injection
AI agents run with their permission checks switched off
AI security review Pro
Claude or Codex reads the findings above, opens the code and tells you which ones are real ship blockers: the few that could cost you your code, your credentials or your customers' data.
At most three findings, each with the path from the attacker's input to the damage, the smallest fix and a test that proves it stays fixed
Every function it names is checked against your repository, so an invented one is marked unverified
AI verdict on the pull request Pro
When a Pro developer pushes a change that adds a risky path, the same AI reads the code on that path and says in the pull request check whether it looks like a real blocker or safe.
Runs only for a paid developer, never automatically for a free account
Saves your team working it out by hand
These checks find risks for your team to review. A flagged path is a statically verified risk, not proof of a working exploit, and they don't replace a penetration test. They don't scan your dependencies for known vulnerabilities or test your running application. See what's free and what's Pro.
From push to proof.
CodeMD works where your team already reviews code.
1
Connect GitHub
Install CodeMD and choose the repositories you want checked.
2
Push a change
CodeMD reviews the change and runs the relevant testing workflow.
3
Merge with evidence
See security findings, test results, and missing coverage before you ship.
A small public demo with both CodeMD checks installed: what the security check blocked and which tests ran. No sign-in needed.
What we found in our own testing.
We took 14 published command-injection advisories for open-source MCP servers, went back to the commit that added each bug, and ran the same free check on it that runs on a pull request.
12 of 14flagged by the check
10on the commit that added the bug
2–15 monthseach bug was in public code before its advisory
These are results from our own test on public code that has since been fixed, with links to each public advisory so you can check them. They show what the check found in these cases; they aren't a guarantee it will find every issue in your code.
CodeMD Pro waitlist
Be first to know when CodeMD Pro launches.
Join the waitlist for AI-powered security reviews, test creation and deeper pull-request insights. Leave an email address, a phone number, or both.
By joining, you agree that CodeMD may contact you about CodeMD Pro. No marketing lists or third-party sharing. See our Privacy Policy.
Free for public repositories. Pro when you want more.
Install the GitHub App and each pull request is checked for new risky paths. Pro adds an AI verdict on each finding, test automation, repository analysis, Testboard and FeatureDashboard.
No credit card. Read-only access to the repositories you choose.
Security Ship Blockers
The Call-Path Security check on every push and pull request. Fails when a change opens a new path from outside input to commands on your server, SQL built from text, eval, HTML written into a page, files served from a variable path, redirects, file writes, git operations or outbound requests. That helps lower the chance of an attacker getting in through a change nobody noticed.
Personal-data checks. Flags passwords and keys, government IDs, health, payment, biometric, identity, contact and location data that a change returns, logs or puts in a URL where outside input can reach it. That helps you catch a possible leak of customer data before it becomes a breach notice or a fine.
Code weakness and AI-attack rules in the Security panel on Analyze Repo. Finds keys and tokens written into the source (AI provider keys too), debug mode left on, weak random numbers used for credentials, plain-text or MD5/SHA-1 passwords, unescaped template output, and redirects taken from a request field, plus AI-specific weaknesses: model output run as code, a shell command or a query, model output written into a page, request or repository text pasted straight into a prompt (the opening for prompt injection), and AI agents run with their permission checks off. No AI is used.
Unlimited public repositories.
Private repositories during early access.
Up to 30 checks per hour per installation, for repositories up to 1,500 MB.
CodeMD Impacted Tests on every pull request, for your existing tests. Runs the tests connected to your change in your repository's own GitHub Actions, which can mean less waiting and fewer Actions minutes than the whole suite, and more chance of catching a bug before customers do. Up to 10 checks an hour per installation, queued behind Pro. A test whose command only AI can work out is skipped.
On-demand repository scans on Analyze Repo: paste a GitHub URL or upload an archive.
Run the tests that map to each commit, from any browser. After an analysis, press Run tests on Analyze Repo; the tests run in your repository's own GitHub Actions and the results go into your Repo Report. 3 runs a day.
Sign in with GitHub to start the trial, no credit card. Checkout and billing are handled securely by Stripe.
Security Ship Blockers, plus
Everything in Free.
AI security review. Claude or Codex reads each risky path, the code weakness and AI-attack rule findings and the personal-data exposures, says whether outside input really reaches the risky call, and lists the real ship blockers (up to three, with the attack path, the smallest fix and a test), in the Security panel on Analyze Repo. It runs only when you press the button.
AI verdict inside the pull request check. When a Pro developer's push adds a risky path, the AI says whether each one is real, safe or unsure, with a fix, right in the check.
First in the queue. Your checks run ahead of every free check and are never turned away when it's busy.
200 checks an hour for each Pro developer, instead of 30 per installation.
Private repositories after early accessComing soon
CodeMD test automation
More from the CodeMD Impacted Tests check: AI finds test commands the free check skips. Up to 20 test checks an hour, ahead of free checks. The pull request check runs your existing tests; it doesn't write new ones.
Creates missing tests on Analyze Repo, from a real caller, or with Claude or Codex when it can't. The free VS Code plugin also creates tests, on your machine.
Files real failures as bugs with the commit, function and test, and auto-fixes them from Analyze Repo.
One account for the VS Code plugin, codemd.dev and the Team FeatureDashboard.
AI features (security review, test generation, auto-fixes, answers) unlock when you upgrade; billing starts that day.