AI coding agents
Security context where code is written.
An agent writing code knows nothing about your findings unless you tell it. Over MCP, Sentrasec hands the agent findings, remediation plans and test plans. The agent makes the change and runs the tests.
Why this matters now
A growing share of production code is written with an agent in the loop. That agent knows the codebase in front of it, but nothing about which dependencies carry known vulnerabilities, which patterns your rules reject, or what the last scan found.
Without that, review is the first point where a security problem surfaces: after the code exists and after the decisions are baked in.
Connected over MCP, the agent can pull the findings for a project, ask why one matters, and receive the remediation plan and test plan. The agent edits the code and runs the tests; a Sentrasec re-scan confirms the finding is gone.
How it connects
The agent talks to Sentrasec over MCP. Sentrasec answers from the same findings, triage decisions and graph as the web app and CLI.
- IDEagent
- MCPtools
- Sentrasecengine
- Knowledge Graphcontext
- Responseinline
Available tools
Standard MCP tool schemas: your client discovers them on connect, and the agent picks the right one from a plain-language request. Fix suggestions are guidance for the agent to apply; Sentrasec never changes your code.
| Tool | What it does |
|---|---|
| security.list_apps | The apps this key can scan and read. Pass one as `app` to security.scan, security.list_scans and security.dast_scan. |
| security.scan | Run a scanner against a registered app: sast, sca, secrets, iac, kspm, apisec, container or dast. Defaults to every code scanner; DAST is left out because it sends live traffic to a running system. Results are deduplicated and saved to your workspace. |
| security.list_findings | List a scan’s findings, optionally filtered to a minimum severity, with file, line and classification. |
| security.get_finding | One finding in full. |
| security.get_scan | Re-read a scan by id: status, who triggered it, and its findings. |
| security.list_scans | An app’s scan history, showing the kind, status and key label behind each run. |
| list_remediations | This app’s remediation plans, newest first: which finding each covers and its state. |
| get_remediation_plan | The full remediation plan for a finding: objective, where to change, impact, steps, constraints, test plan and definition of done. It contains no code and no diff; your agent makes the change. |
| get_test_plan | Just the test plan: numbered test obligations, who owns each, and the rule for passing. |
| request_verification | After the change is pushed, ask Sentrasec to re-scan that commit and check the test plan’s obligations against it. |
| security.explain | Explain a vulnerability by its knowledge-graph node id: description, CWE, OWASP mapping and remediation guidance. |
| security.explain_iac_finding | Why an IaC finding was raised: the resource, the rule, the value observed, what was expected, and how risk was scored. |
| security.scan_local | Scan a local directory in-process, without an app or credentials. Results stay in the session. Prefer security.scan once an app is registered. |
| security.list_findings_local | List findings from a security.scan_local run. |
| security.get_finding_local | One finding from a security.scan_local run, in full. |
| security.fix | For local scans only: propose a tier-1 autofix diff for a finding. Dry run by default, so your agent can review it; apply=true writes the diff to your working copy. |
| security.dast_scan | Probe a URL that is not yet registered as an app. Prefer security.scan with kind=dast when the app exists. |
| security.list_rules | List the detection rules available in the engine. |
| security.list_languages | List the languages the scanners support, with tier and parser status. |
| security.version | Return the running sentrasec-mcp server version. |
| security.issue_cert | Issue a scan certificate covering a set of scans. It is unsigned today, so treat it as a scan summary, not an attestation a third party can verify. |
| security.verify_cert | Look up a certificate by id, returning its grade and validity. |
Supported clients
Any MCP-capable client works. These are the ones with documented setup guides.
Claude Code
Add the server once; the tools are available in every session.
Cursor
Add the MCP server in settings and the agent can call the tools inline.
Codex
Register the server in the MCP configuration and ask for findings in plain language.
Windsurf
Add the server to the MCP configuration to bring findings into the agent.
GitHub Copilot
Connect through the agent MCP configuration to bring findings into chat.
VS Code
Built-in MCP support connects the server to the editor agent.
$ claude mcp add sentrasec -- sentrasec mcpThe server acts as your authenticated user, so the agent can only read findings you can read. Per-client setup is on the MCP site.
Security as a capability, not a gate
Full MCP reference, tool schemas and per-client setup.