tracer.

Application security / Built for AI development

Find the flaw. Prove the risk.

Tracer connects what your code says with what your application actually does. Find vulnerabilities, correlate evidence, and verify the fix.

Runs as a CLI, an MCP server for coding agents, and a self-hosted web UI.

Example trace with illustrative data

A database migration creates the table public.profiles without row level security. An anonymous request to the live application's REST endpoint for profiles returns HTTP 200 with readable rows. Both observations point at the same resource, table public.profiles, so Tracer reports the finding as confirmed, critical.

Finding vulnerabilities is easy.Knowing which ones matter is harder.

Static analysis identifies potential weaknesses. Runtime testing observes application behavior. Viewed separately, each tells only part of the story.

From the source

A static finding without runtime context

src/lib/payments.ts:4
const stripe = new Stripe("‹REDACTED sha256:5d41a07c93be …9f2c›");

Stripe live secret key hardcoded in source

What the code can't tell you

  • Is this file part of what's deployed?
  • Can a visitor actually retrieve the key?

Source onlyFound in source. Not reproduced live.

From the running application

A runtime observation without a known source

GET https://app.example.com/api/export
200 content-type: application/json, 18,342 bytes

/api/export returns data to an unauthenticated request

What the response can't tell you

  • Which handler serves this route?
  • Is the auth check missing, or misconfigured?

Live onlyObserved live. No source origin located.

Tracer connects the evidence.

When a source finding and a live observation resolve to the same resource, the finding is confirmed. When they don't, Tracer tells you which half is missing.

From observationto confirmation.

One finding, followed from a line in a migration to a confirmed, fixable issue. The example data is illustrative; the steps are the ones the engine runs.

  1. 1 of 5

    Source analysis identifies a possible weakness

    The migration creates public.profiles and never enables row level security. On its own, this is a hypothesis: the table might not be served, or access might be restricted somewhere Tracer can't see.

    Rule
    rls-missing-policy
    Status
    Source only, high
  2. 2 of 5

    A runtime check observes compatible behavior

    With the public anon key from the deployed bundle and no user session, a request to /rest/v1/profiles returns 200 and readable rows. Tracer reads one row to confirm the exposure and records the count and column names, never the values.

    Rule
    rls-exposed-table
    Status
    Live only, critical
  3. 3 of 5

    The engine evaluates the relationship

    Each observation carries a resource identity. Both resolve to table:public.profiles, and a correlation rule pairs these two checks. Matching is by identity, not similarity: a different table, or a header on a different route, would not join.

    Join key
    table:public.profiles
    Pair
    rls-missing-policy + rls-exposed-table
  4. 4 of 5

    A supported confirmation is reported

    Source and live checks agree, so the finding is CONFIRMED and sorts first. Its severity is critical: the missing policy alone was high, and the live probe shows the table is reachable from the internet.

    Rule
    rls-missing-policy-exposed
    Status
    CONFIRMED, critical
  5. 5 of 5

    A remediation recommendation follows

    The fix prompt carries corrective SQL scoped to user_id, the column with a foreign key to auth.users. Your coding agent applies it. After deploying, rescan re-runs the same checks and reports resolved, improved, unchanged, or inconclusive.

    Fix
    Enable RLS, add owner policies
    Verify
    rescan the finding

One workflow.From detection to verification.

01. Inspect

Static source-code analysis

Identify suspicious code patterns, exposed secrets, and security configuration problems in the repository and its infrastructure files.

  • Hardcoded secrets and committed .env files
  • Supabase migrations without row level security
  • Permissive Firebase rules and open storage buckets
  • Unprotected API routes and missing security headers
  • SQL injection, unsafe eval and dangerous CORS, via Semgrep
Static pass over ./acme-portal
  • supabase/migrations/20240101000000_init.sqlrls-missing-policy2 tables
  • src/lib/payments.tshardcoded-secret‹REDACTED …9f2c›
  • firestore.rulesno finding
  • src/app/api/export/route.tsno finding
  • netlify.tomlno finding

Secret values never leave the scan. Findings record a fingerprint and the last four characters.

02. Observe

Dynamic application analysis

Inspect an authorized running application to detect weaknesses you can observe from outside. Probes do the minimum needed to confirm, and stop.

  • Secrets in the served JavaScript bundle
  • Anonymous reads through Supabase's REST API
  • Open Firebase databases and storage buckets
  • Routes that return data without a session
  • Reflected XSS, and IDOR enumeration only on opt-in
Probes against https://app.example.com

AuthorizationAttested by you before any probe ran

RequestStatusObservation
GET /assets/index-4f1c2a.js200Secret pattern in served bundle
GET /rest/v1/profiles2001,247 rows, anonymous
GET /api/export200JSON with no session
GET /200Headers checked
idor-object-accessskippedAggressive probe, opt-in only

03. Correlate

Evidence correlation

Connect compatible static and runtime observations to establish a stronger finding. Anything left unmatched is still reported, with the side that's missing.

  • Joins on secret fingerprints, table names, routes and buckets
  • Each pairing is an explicit rule, never a similarity guess
  • CONFIRMED findings sort above equal-severity ones
Joining on resource identity
SourceMatchLiveResult
secret:sha256:5d41a07c93be secret:sha256:5d41a07c93be CONFIRMED
table:public.profiles table:public.profiles CONFIRMED
table:public.audit_logs none Source only
none route:/api/export Live only

04. Verify

Remediation verification

Re-run the checks behind a finding to determine whether the issue is still observable, from the CLI, the web UI, or your agent.

  • Rescan one finding, or the whole target
  • Outcomes: resolved, improved, unchanged, or inconclusive
  • Exit codes gate CI with --fail-on
Rescanning one finding

rescan(finding_id: "7c1e9a40d2b35f81")

Before

CONFIRMEDtable:public.profiles

After the fix deploys

resolved

resolved
The checks re-ran cleanly and no finding for the resource remains.
improved
One side remains, e.g. a fixed migration that isn't deployed yet.
unchanged
The finding is still present at the same strength.
inconclusive
A check couldn't complete, or an ignore rule hides it. Missing evidence is not a fix.

Matched by resource, not by line, so a fix that moves code still resolves correctly.

Security findingsyou can investigate.

Every finding opens on its evidence: the line in the repository, the request against the deployment, the resource they share, and a fix prompt your agent can apply.

Illustrative preview of the self-hosted web UI, tracer serve, with fixture data. Not a live scan; Rescan runs in the real app.

tracer. Traces acme-portal Scan complete

CONFIRMED. Source and live checks agree.

2 confirmed1 live only1 source only

Repo
./acme-portal
URL
https://app.example.com
Authorization
Attested by you
  • Not run1 aggressive probe (idor-object-access). Opt in with --aggressive.
  • SkippedSemgrep isn't installed, so SQL injection, unsafe deserialization and CORS checks didn't run.
CONFIRMEDcriticalCWE-798

Stripe live secret key hardcoded in the repo AND served in the live bundle (src/lib/payments.ts)

hardcoded-secret-exposedid e04b7d19c6a2f358

Source

src/lib/payments.ts:4

const stripe = new Stripe("‹REDACTED sha256:5d41a07c93be …9f2c›");
Live200

GET https://app.example.com/assets/index-4f1c2a.js

…apiVersion:"2024-06-20"}),Ks=new Stripe("‹REDACTED sha256:5d41a07c93be …9f2c›")…

Retrieved anonymously with a plain GET — no authentication and no evasion. The stripe live secret key (sha256:5d41a07c93be (…9f2c, 107 chars)) is present in a JavaScript bundle served to the browser, so it is readable by anyone who loads the page.

secret:sha256:5d41a07c93beCONFIRMED

Source and live checks agree.

Fix prompt

Written for your coding agent. Apply it, then rescan.

CONFIRMED: a Stripe live secret key (sha256:5d41a07c93be …9f2c) is committed in src/lib/payments.ts AND served to every visitor in the live JavaScript bundle.

1. Rotate the key in the Stripe dashboard first. Removing it from the code does not revoke it.
2. Move the call that needs it to a server route; read the key from an environment variable that is never exposed to the client bundle.
3. Remove the literal from src/lib/payments.ts and rebuild, so the new bundle no longer contains it.
4. Rescan. The bundle check must come back clean for the new deployment.

Security belongs in thedevelopment workflow.

Run Tracer where you build. Inspect findings, generate remediation guidance, and verify changes without switching tools.

Tracer CLI, illustrative output
$ tracer scan repo ./acme-portal --url https://app.example.com --authorized-by "you"      .-''''-.     ▪      .' .----. '.      tracer. v0.1.0  -(--| ▌  ▌ |--)-    Tracing source and live surfaces…    '. '----' .'  ▪     ▪   '-....-'          ✓ Static analysis       7 checks · 3 signals  ✓ Dynamic testing       6 checks · 3 signals  ● Correlating findings  2 signals matched  ────────────────────────────────────────────────────────────────   CONFIRMED   ▮▮▮▮ CRITICAL hardcoded-secret-exposed  Stripe live secret key hardcoded in the repo AND served in the live bundle  ────────────────────────────────────────────────────────────────   CONFIRMED   ▮▮▮▮ CRITICAL rls-missing-policy-exposed  Table public.profiles has no row level security and is readable by anyone  SOURCE  supabase/migrations/20240101000000_init.sql:8    create table public.profiles ( id uuid primary key default gen_random_uuid(), …  LIVE    GET https://app.example.com/rest/v1/profiles → 200    HTTP 200 · 1247 row(s) readable. Columns: id, user_id, email, full_name.    Requested anonymously with the public anon key. Values not recorded.  ↳ CONFIRMED · Source and live checks agree on table:public.profiles  ────────────────────────────────────────────────────────────────  LIVE ONLY    ▮▮▮▯ HIGH     unauthenticated-route  /api/export returns data to an unauthenticated request  ────────────────────────────────────────────────────────────────  SOURCE ONLY  ▮▯▯▯ LOW      rls-missing-policy  Table public.audit_logs has row level security enabled but no policies  (evidence shown for one finding; the rest is shortened on this page)  ────────────────────────────────────────────────────────────────  CONFIRMED · Source and live checks agree.  2 CONFIRMED  ·  1 source only  ·  1 live only  2 critical  ·  1 high  ·  1 low  1 aggressive probe(s) not run (opt in with --aggressive): idor-object-access  warning Semgrep is not installed, so code-layer checks (SQL injection, unsafe  deserialization, dangerous CORS) were skipped. Install it with `pipx install semgrep`.
Coding agent with Tracer's MCP server, illustrative session
  1. You

    Explain the confirmed RLS finding.

  2. Tool call

    tracer.get_fix_prompt(finding_id: "7c1e9a40d2b35f81")

  3. Agent

    public.profiles is created in 20240101000000_init.sql without row level security. A live, anonymous GET /rest/v1/profiles using only the public anon key returned 200 with 1,247 readable rows. Both observations resolve to table:public.profiles, so Tracer reports it as CONFIRMED and critical.

    The fix enables RLS and adds select, insert and update policies scoped to user_id, which references auth.users. Treat the table's contents as already disclosed.

  4. You

    Apply it, then verify.

  5. Agent edit

    Created supabase/migrations/20261013090000_profiles_rls.sql

    Made by your coding agent, not by Tracer.

  6. Tool call, after deploy

    tracer.rescan(finding_id: "7c1e9a40d2b35f81")

  7. Result

    resolved No finding for table:public.profiles remains, and both checks re-ran cleanly.

  • CLI

    Available

    $ tracer scan repo . --url https://app.example.com

    A readable report, or SARIF, JSON, Markdown and HTML. Exit codes gate CI with --fail-on.

  • MCP

    Available

    $ tracer mcp

    Four tools for your agent: scan_repo, scan_url, get_fix_prompt and rescan. Over stdio, or HTTP from the web server.

  • Web

    Self-hosted UI available. Hosted platform in development.

    $ tracer serve

    UI, REST API and MCP on one local port. Loopback by default; scans live in memory and are gone on restart.

Your tools.Your workflow.

Tracer speaks MCP and SARIF, so it fits the agents and pipelines you already run. Status reflects what ships today.

Works today

  • Claude Code

    MCP server

    Available

    scan_repo, scan_url, get_fix_prompt and rescan, over stdio or HTTP.

  • Cursor

    MCP server

    Available

    The same MCP configuration as Claude Code.

  • GitHub

    SARIF and CI exit codes

    Available

    Upload SARIF to code scanning and gate builds with --fail-on. Connected repositories are planned.

tracer.

MCP server SARIF CLI

Coming

  • ChatGPT

    App integration

    In development

    Not available yet.

  • Codex

    MCP server

    Planned

    Codex supports MCP servers; a tested Tracer configuration isn't published yet.

  • Available
  • In development
  • Planned

Built for builders.Useful to the people responsible for shipping.

Developers

Understand security findings without leaving the development workflow.

You asked an agent to add Supabase auth. Tracer checks what actually shipped and hands the agent a fix prompt in the same session.

In your agent
scan_repo → get_fix_prompt → rescan

Engineering teams

Review application security evidence and prioritize remediation.

CONFIRMED findings sort first, so review starts with what's proven reachable. SARIF lands in GitHub code scanning next to the code.

In CI
tracer scan repo . --fail-on high
tracer scan repo . --format sarif > tracer.sarif

Consultants and agencies

Document security reviews and generate client handoff reports.

Before handing a build to a client, record who authorized the scan, what was checked, what was skipped and why, and the residual risk.

At handoff
tracer handoff ./acme-portal \
  --url https://app.example.com \
  --authorized-by "you" --client "Acme Corp" \
  --out handoff.md

Security tools shouldbe inspectable.

Tracer began as an open-source, local-first scanner for developers, and the scanning engine is MIT-licensed. The hosted platform now in development builds on that same engine. You should be able to read exactly why a finding was raised.

Inspectable detection rules
Each check is a readable module with a vulnerable fixture and a documentation page. Code-pattern checks run Semgrep rules you can open and read.
Evidence-backed findings
Every finding carries what was observed: the file and line, the request and the response. Secrets appear only as a fingerprint and their last four characters.
Reproducible checks
Finding ids are deterministic, so the same issue keeps its id across scans, and rescan re-runs the exact scan that produced a finding.
Developer-first integrations
SARIF for code scanning, JSON and Markdown for tooling, exit codes for CI, and an MCP server for coding agents.
Safe by default
Live probes need an authorization attestation, aggressive probes are opt-in, and a probe does the minimum needed to confirm a finding.

Quickstart

The core commands. Probing a live URL always requires --authorized-by; scans that only read the repository don't.

Scan a repository, then correlate it with the deployment
$ tracer scan repo ./my-app
$ tracer scan repo ./my-app --url https://my-app.example.com --authorized-by "Your Name"
Register the MCP server with your editor
{
  "mcpServers": {
    "tracer": { "command": "tracer", "args": ["mcp"] }
  }
}
Run the self-hosted UI, API and MCP endpoint
$ tracer serve        # http://127.0.0.1:7341

Don't just find it. Understand it.

Security findings supported by evidence. Built for modern software development.

Explore Tracer