AppSecGate MCP server
Commit Nov 7, 2026Enables AI agents to perform local-first security scanning on files, diffs, and entire projects without sending code to external servers.
Also known as moondoom/appsecgate-mcp
- Stars
- 2
- Tools
- 17
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 7, 2026
- License
- MIT
What is AppSecGate MCP?
AppSecGate enables AI agents to perform local-first security scanning on files, diffs, and entire projects without sending code to external servers. It exposes 17 tools, such as scan_file, scan_diff, scan_project.
Summary generated from the project's README
About AppSecGate MCP server
Performing local-first security scans on files, diffs, and entire projects is what appsecgate-mcp enables for developers using AI coding agents. This server integrates directly into environments like Claude Code, Cursor, Cline, or Continue to provide security feedback without sending source code to external servers. It operates as a hardened Docker container with no network access, ensuring that all scanning remains offline and private.
Developers use specific tools to manage their security posture, such as scan_file for individual files, scan_diff for git changes, and scan_project for full repository audits. The server provides specialized functions like get_fix_context to give agents the necessary information to repair vulnerabilities and verify_fix to confirm that a patch successfully addressed a finding. By using Semgrep for SAST, Gitleaks or TruffleHog for secrets, and Trivy for SCA, it brings professional-grade scanning directly into the agentic workflow.
What you can do with it
- 1Scan git changes before committingAn AI agent can scan staged and unstaged changes in a repository to catch vulnerabilities before they are merged.
- 2Provide context for automated repairsThe agent can retrieve the specific code window and rule documentation needed to fix a security finding.
- 3Verify security patchesAfter applying a fix, the agent can rescan the file to ensure the vulnerability is no longer present.
- 4Maintain offline security workflowsTeams can run security audits on their local machines without any code leaving the environment or making network calls.
Written from the project's README. Every use case is checked against a quote from it.
AppSecGate MCP tools 17
Detected from the READMEscan_filescan_diffscan_projectget_scan_findingsread_project_filecheck_gateget_fix_contextget_fix_context_batchverify_fixverify_fix_batchquery_findings_dbexport_snapshot
Show all 17 toolsShow fewer tools
get_configupdate_configsubmit_feedbackget_scan_historysuppress_finding
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachablemoondoom/appsecgate-mcp is public and reachablegithub.com/moondoom/appsecgate-mcp ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-07github.com/moondoom/appsecgate-mcp ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model17 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubAppSecGate MCP
Local-first security scanning for AI coding agents.
AI agents ship code faster than security can review it. AppSecGate is an MCP server that scans files, diffs, and whole projects right inside Claude Code, Cursor, Cline, or Continue — offline, without sending your code anywhere. Findings are normalized, deduplicated by fingerprint, tracked between scans, and returned to the agent as fix contexts.
Security should not be a drag. It should be invisible to the developer and transparent to security.
Read: Making AI agents paranoid: will MCP replace AppSec? (Habr, RU)
When to use / When NOT to use
Use it when:
- You build with an AI agent and want every diff scanned before it lands.
- You want findings tracked across scans (deduplicated by fingerprint), not a pile of reports.
- You want to stay offline: no network calls, no code leaves your machine, and the server itself never calls an LLM.
- You want your agent to see the context needed to fix a finding (
get_fix_context) and to verify the fix later (verify_fix).
Do NOT use it when:
- You need large-scale GUI triage, RBAC, team dashboards, or ASOC sync. Local scanning, centralized policy, and classic CI blocking are already implemented; dashboards and ASOC integration are not.
Demo
Quick Install
The server runs as a Docker container. Your project is mounted read-only
(:ro), the container has no network access (--network=none), and scans
never write to your source tree.
docker pull moondoom/appsecgate-mcp:0.1.0
Claude Code
Add to <project>/.mcp.json:
{
"mcpServers": {
"appsecgate": {
"command": "docker",
"args": [
"run", "--rm", "-i", "--network=none", "--read-only",
"--tmpfs=/tmp:size=100M,mode=1777", "--cap-drop=ALL",
"--security-opt=no-new-privileges:true",
"-v", ".:/workspace:ro",
"-v", "appsecgate-cache:/home/appsecgate/.cache",
"-v", "appsecgate-data:/home/appsecgate/.appsecgate-mcp",
"moondoom/appsecgate-mcp:0.1.0",
"--transport", "stdio", "--project", "/workspace"
]
}
}
}
Or register it directly:
claude mcp add appsecgate -- docker run --rm -i --network=none --read-only \
--tmpfs=/tmp:size=100M,mode=1777 --cap-drop=ALL \
--security-opt=no-new-privileges:true \
-v .:/workspace:ro \
-v appsecgate-cache:/home/appsecgate/.cache \
-v appsecgate-data:/home/appsecgate/.appsecgate-mcp \
moondoom/appsecgate-mcp:0.1.0 --transport stdio --project /workspace
Cursor
Add to <project>/.cursor/mcp.json:
{
"mcpServers": {
"appsecgate": {
"command": "docker",
"args": [
"run", "--rm", "-i", "--network=none", "--read-only",
"--tmpfs=/tmp:size=100M,mode=1777", "--cap-drop=ALL",
"--security-opt=no-new-privileges:true",
"-v", ".:/workspace:ro",
"-v", "appsecgate-cache:/home/appsecgate/.cache",
"-v", "appsecgate-data:/home/appsecgate/.appsecgate-mcp",
"moondoom/appsecgate-mcp:0.1.0",
"--transport", "stdio", "--project", "/workspace"
]
}
}
}
Cline
Cline (v4.x) reads MCP servers from the shared settings file
~/.cline/data/settings/cline_mcp_settings.json. Add the appsecgate server:
{
"mcpServers": {
"appsecgate": {
"type": "stdio",
"command": "docker",
"args": [
"run", "--rm", "-i", "--network=none", "--read-only",
"--tmpfs=/tmp:size=100M,mode=1777", "--cap-drop=ALL",
"--security-opt=no-new-privileges:true",
"-v", "/absolute/path/to/your/project:/workspace:ro",
"-v", "appsecgate-cache:/home/appsecgate/.cache",
"-v", "appsecgate-data:/home/appsecgate/.appsecgate-mcp",
"moondoom/appsecgate-mcp:0.1.0",
"--transport", "stdio", "--project", "/workspace"
],
"cwd": "/absolute/path/to/your/project",
"env": {},
"disabled": false,
"timeout": 600
}
}
}
For a shared (user-level) config use an absolute project path in -v and
cwd; a relative .:/workspace:ro works only in a per-project .mcp.json.
Continue
Add to <project>/.continue/mcpServers/appsecgate.yaml:
name: appsecgate
version: 0.1.0
schema: v1
mcpServers:
- name: appsecgate
type: stdio
command: docker
args:
- run
- --rm
- -i
- --network=none
- --read-only
- --tmpfs=/tmp:size=100M,mode=1777
- --cap-drop=ALL
- --security-opt=no-new-privileges:true
- -v
- .:/workspace:ro
- -v
- appsecgate-cache:/home/appsecgate/.cache
- -v
- appsecgate-data:/home/appsecgate/.appsecgate-mcp
- moondoom/appsecgate-mcp:0.1.0
- --transport
- stdio
- --project
- /workspace
connectionTimeout: 600000
requestOptions:
timeout: 600000
Verify the install: restart the client, then run scan_file on any file; tools
appear with the mcp__appsecgate__ prefix. Full instructions, edge cases, and
diagnostics: docs/deploy-agents.md.
Agent mode (coming in 0.1.8)
The read-only agent autonomously triages the complete current inventory instead of only the top 10 findings. It chooses scanner context, bounded file reads, historical human-confirmed verdicts, and CVE data through tools. A deterministic controller tracks pending IDs, detects loops, restarts from checkpoints, compacts context, and preserves a redacted verdict journal. Synchronization uses the existing edge outbox; master MCP is read-only. The agent never modifies the project or applies patches.
The LLM runs outside the MCP server and is configured through an OpenAI-compatible endpoint; the server's no-LLM invariant is unchanged.
appsecgate-mcp agent \
--mode=ro --scan=full --policy=warn --output-format=markdown \
--project . \
--llm-url http://<llm-host>/v1 --llm-model <model>
See the TP-0001 experiment report for model benchmarks and aggregate results from 100 repositories.
MCP tools (17)
| Tool | What it does |
|---|---|
scan_file |
Scan a single file and return normalized findings. |
scan_diff |
Scan staged + unstaged + untracked changes in the current git repository. |
scan_project |
Scan the entire project (no arguments) — reconciliation report: counts + TOP-10 + requires_confirmation. |
get_scan_findings |
Return a deterministic page of findings from the latest scan, including the full inventory hash. |
read_project_file |
Read a bounded, redacted line range from a file inside the project. |
check_gate |
Evaluate the latest scan under an advisory policy and return a GateSummary with a recommended action. |
get_fix_context |
Return the ContextPack for a finding (finding + code window + rule doc). |
get_fix_context_batch |
Return ContextPacks for multiple findings (limit 20). |
verify_fix |
Verify a single finding: rescan the file and match by (logical_id). |
verify_fix_batch |
Verify up to 20 findings; each file is rescanned once. |
query_findings_db |
Read-only SELECT on public views (v_findings, v_summary, v_projects). |
export_snapshot |
Export a snapshot of findings in ndjson format (parquet — Phase 2). |
get_config |
Return the active (merged) session configuration. |
update_config |
Deep-merge updates into .appsecgate-mcp/config.yaml and reload the session. |
submit_feedback |
Submit feedback on a finding (FALSE_POSITIVE / CONFIRMED / WONT_FIX). |
get_scan_history |
Return the shown findings log (shown_findings.db) with verification attempts. |
suppress_finding |
Create or remove a local suppression (action: suppress|unsuppress; does not affect CI gate in Phase 1). |
Scanners (Phase 1)
| Category | Scanners | Notes |
|---|---|---|
| SAST | Semgrep | Bundled offline rules (Rules Bundle) — no registry access needed |
| Secrets | Gitleaks, TruffleHog | Values redacted before they are stored |
| SCA | Trivy | Vulnerability DB seeded at image build time |
Every scanner runs as a separate subprocess (no shell). Planned for later phases: Bandit, ESLint (SAST); OSV-Scanner, Dependency-Track (SCA); Checkov, KICS (IaC); OWASP ZAP (DAST).
Security model
- Read-only project. The project is mounted
:ro; the server never writes to source files. - Offline by default.
network_mode: none— no outbound network calls, ever. - No LLM calls. The fix is generated by your agent's own model; the server only ships context data, never prompts.
- Secrets redacted before storage. At write time, not at read time.
- Hardened container. Non-root UID 1000, read-only rootfs,
cap-drop=ALL,no-new-privileges.
Configuration
Per-project settings live in .appsecgate-mcp/config.yaml and are deep-merged over
the image defaults (/etc/appsecgate/default.yaml):
version: 1
stack:
languages: [python, typescript]
frameworks: [fastapi, react]
scanners:
sast:
enabled: true
tool: semgrep
rulesets: [] # semgrep rules come from the local Rules Bundle
secrets:
enabled: true
tool: gitleaks
sca:
enabled: true
tool: trivy
severity_threshold: HIGH
gates:
local:
warn_on: [CRITICAL, HIGH, MEDIUM]
block_on: [] # never block locally
ci:
block_on: [CRITICAL, HIGH]
warn_on: [MEDIUM]
block_on_unresolved_local: true # "seen but not fixed" (Phase 2)
diff_base_ref: origin/main # base for `appsecgate ci`
Optional: user-downloaded Semgrep rules
The image ships only rules that AppSecGate may redistribute. You can download additional rules yourself; they are stored in the user data directory, not in the project, image, or bundled rules. List the pinned sources and licenses:
appsecgate-mcp rules list
appsecgate-mcp rules install \
--source semgrep --source trailofbits --profile agent300 --accept-licenses
For a custom selection, replace --profile agent300 with, for example,
--languages python,javascript,go. The agent300 profile keeps Python,
JavaScript, TypeScript, Go, Java, Ruby, PHP, and configuration/IaC rules while
excluding unrelated languages and test fixtures. Downloading happens only in
this host command; MCP scanning remains offline. No image rebuild is needed.
Set APPSECGATE_RULES_DISTRIBUTION_MODE=commercial on the MCP container and
mount the same ~/.appsecgate-mcp host directory at
/home/appsecgate/.appsecgate-mcp. With a named appsecgate-data volume, run
the install command in a separate networked container using that volume. The
commercial setting means that you accept responsibility for the selected
licenses; it is not an AppSecGate commercial license. In particular, Semgrep
rules are for internal use and must not be redistributed.
Review rules/.external-rules.json for revisions, hashes, selected files, and
duplicates. Remove a source with appsecgate-mcp rules remove --source semgrep.
A changed ruleset also changes scan results and ruleset_hash.
Optional: centralized CI and master sync
Local MCP scanning stays offline and warn-only. For shared findings, policy synchronization, and blocking CI scans, use the separate master + edge setup described in docs/deploy-master.md.
Build from source
make setup # create the uv environment
make check # lint (ruff) + tests (pytest)
make docker-build # build moondoom/appsecgate-mcp:0.1.0
Documentation
- Agent deployment guide — all four clients, verification steps, troubleshooting
- Master + edge deployment — master server, sync, CI gate
- Architecture
- E2E test plan
- E2E report
- TP-0001 Stage 2 test plan
- TP-0001 Stage 2 experiment report — model benchmarks and aggregate findings
- Product vision
- Open tasks
- Article on Habr (RU) — "Making AI agents paranoid: will MCP replace AppSec?"
License
MIT — see LICENSE. Bundled scanner rules are third-party (see third-party notices).
Russian version: README.ru.md.
AppSecGate MCP questions
What tools does AppSecGate MCP provide?
It exposes 17 tools, including scan_file, scan_diff, scan_project, get_scan_findings, read_project_file.
Can I use this while staying offline?
Yes, the server is designed to run offline with no network access, ensuring no code leaves your machine.
Does the server modify my source code?
No, the project is mounted as read-only and the server never writes to your source tree.