English

Using Aikido MCP

The tools the Aikido MCP server exposes, and example prompts for asking your agent to use them.

Once the MCP server is installed, you don't call these tools yourself. You ask in plain language and your agent picks the right one.

Available tools

aikido_scan_paths

Scans files that are saved on disk for vulnerabilities (SAST), hardcoded secrets and IaC misconfigurations. This is the preferred scan tool.

  • Provide the files to scan in paths, absolute or relative to root. Directories are rejected, so list each file
  • Set root to the workspace or repository root so findings are reported relative to it. Paths that are missing or outside root are skipped and listed in skipped_paths
  • Optionally pass repository_name when the files come from a repository, so they are linked to that repository in Aikido

aikido_full_scan

Scans code that is passed in as file content, for the cases where the code is not on disk yet: generated but unsaved code, a pasted snippet, or a proposed change. Use aikido_scan_paths whenever the files do exist on disk.

  • Provide each file in files_to_scan as a relativeFilePath and its content

aikido_oss_dependency_scan

Scans dependency manifests and lockfiles that are saved on disk for vulnerable, outdated or malicious packages.

  • Provide the manifests and lockfiles to scan in paths, for example package.json and package-lock.json, requirements.txt, go.mod and go.sum, or pom.xml. Directories are rejected
  • Set root to the workspace or repository root so findings are reported relative to it
  • Optionally pass repository_name when the files come from a repository

aikido_issues_list

Fetches security issues from your Aikido feed.

  • Filter by scope: repo_name, cloud_name, vm_name, domain_name, container_name, workspace_name or team_name
  • Optionally narrow a repo scope to a single branch with repo_branch_name
  • Pick one or more issue types: sast, leaked_secret, iac, open_source, cloud, cloud_instance, docker_container, malware, eol, mobile, surface_monitoring, scm_security, license, ai_pentest, ai_code_analysis
  • Filter by labels to only return issues on repositories or container images carrying any of the given labels
  • Filter by severity to only return issues of a certain severity or above a given threshold
  • Page through results with page (zero-based)
  • Filter on SLA by choosing the following filters: out_of_sla or sla_due_soon
  • Returns each issue with title, type, severity, and remediation steps

Looking up a single issue

Pass issue_id to fetch one issue in detail. It comes back with more than the list view shows:

  • score_reasons: every reason the issue got its severity, each with a readable explanation and how much it moved the score, for example a known exploit in the wild
  • pentest: for AI pentest findings, the attack analysis with its summary, risk, reproduction steps, root cause and remediation
  • licenses: for license issues, the detected license with its legal risk and why

aikido_workspaces_list

Lists the Aikido workspaces you have access to. Your agent uses it to find the exact workspace name before filtering issues by workspace_name.

aikido_ignore_issue

Ignores a security issue in the feed. Requires issue_id and a reason.

aikido_unignore_issue

Unignores a previously ignored issue. Requires issue_id and a reason.

aikido_snooze_issue

Temporarily hides a security issue in the feed. Requires issue_id, a reason and a duration: 1_day, 1_week, 1_month, 3_months, 6_months or 1_year.

aikido_unsnooze_issue

Unsnoozes a previously snoozed issue. Requires issue_id.

aikido_adjust_issue_severity

Raises or lowers the severity of a security issue. Requires issue_id, the new adjusted_severity (critical, high, medium or low) and a reason.

aikido_login

Starts the Aikido sign-in flow; returns sign-in URL or confirms you're already signed in.

  • Sign-in redirects you to your account's region automatically. Only set region (eu, us, me, au) if you land in the wrong one
  • Use force_reauth to switch accounts or sign in again

Not all MCP tools are enabled by default. Admins can enable them on the permissions page for everyone

Example prompts

Scanning code

  • "Use Aikido to scan this file for security issues"
  • "Run an Aikido scan on my staged changes to check for secrets before I commit"
  • "Scan the files I just edited with Aikido and link them to the payments-api repo"

Reviewing issues by repo

  • "Show me all critical Aikido issues in payments-api"
  • "List any leaked secrets in frontend-web from Aikido"
  • "What open source vulnerabilities does Aikido see in api-gateway?"
  • "Show SAST and IaC issues in infra-core from Aikido"

Reviewing issues by cloud, VM, or container

  • "List all Aikido cloud issues in prod-aws"
  • "Show malware findings on web-server-01 from Aikido"
  • "What end-of-life software is running in the nginx-proxy container per Aikido?"
  • "Show me surface monitoring issues for example.com in Aikido"

Digging into one issue

  • "Give me the full details of Aikido issue 12345"
  • "Why is Aikido issue 12345 rated critical? Is there a known exploit?"
  • "Walk me through the reproduction steps for the Aikido pentest finding 12345"

Managing issues

  • "Snooze Aikido issue 12345 for a month, we're fixing it in the next release"
  • "Lower the severity of Aikido issue 12345 to low, the endpoint is internal only"
  • "Unignore Aikido issue 12345, the mitigation was removed"
  • "Which Aikido workspaces do I have access to?"

Combined workflows

  • "Use Aikido to scan my current changes, then show existing critical issues in the same repo"
  • "Check this PR with Aikido and compare against open SAST issues in the repo"

Controlling when scans run

Aikido does not decide when to scan. Your agent does. A scan happens when the agent calls aikido_scan_paths.

So anything you tell the agent about scanning takes effect straight away, for the rest of the session.

Pause scanning

  • "Don't run any Aikido scans until I ask for one."
  • "Pause Aikido scanning while I iterate on this."
  • "Skip the Aikido scan for now, I'll ask for one before we commit."

Scan once at the end instead of after every edit

  • "Don't scan after each edit. Run one Aikido scan when the whole task is done."
  • "Run an Aikido scan on my staged changes right before I commit, not before that."

Keep scans small

Scan time scales with how much code goes in, so narrowing the scope is usually faster than switching scanning off.

  • "Only scan the files you changed, not the whole directory."
  • "Scan src/payments/ only."
  • Prefer aikido_scan_paths over aikido_full_scan. It reads the files from disk instead of passing their content through the conversation.

Make it your default

A prompt instruction lasts for one session. To change the default for every session, put the same instruction in your agent's rules file (CLAUDE.md, .cursor/rules, AGENTS.md, and so on):

Run an Aikido scan only when I explicitly ask for one, or before a commit.