Sigrid Axis MCP tools reference

The Sigrid Axis MCP server gives an agent one Guardrails check that reads the code in your working tree, and tools that read what Sigrid Core found in a published system or write triage decisions back to it.

The skills call these tools for you. You need this page when you prompt the tools directly, restrict which ones a session sees, or debug a connection. For connecting the server, see installation.

How tool names appear

This page writes the tools in dotted form, such as guardrails.quality_check. Clients display them differently. Claude Code shows guardrails_quality_check, and with the plugin it adds the prefix mcp__plugin_axis_axis__, which is also the prefix to use in a permission allowlist. Match on the tool, not on the exact string.

Every tool except guardrails.quality_check takes a customer and a system, the names in your Sigrid URL. The parameters below are the ones that shape the result.

Tools at a glance

Guardrails

guardrails.quality_check checks one file for maintainability and security issues. It returns the maintainability guidelines the file violates, with a severity and a line range per unit, plus a separate list of security findings. Both lists empty means a pass. It does not return a rating.

• filename: the original file name, including the extension, so the right language rules apply.
• code_snippet: the file’s contents, exactly as written to disk, not a diff or a summary. For Java, pass a complete class that compiles, because the analyzer compiles it.

See supported technologies.

Maintainability

maintainability.get_ratings returns the current maintainability ratings, on a scale from 0.5 to 5.5 stars, where 3.0 is the market average and 4.0 is the target for new development.

• component: set to true for ratings per component instead of for the whole system.
• technology: set to true to add statistics per technology.

maintainability.get_findings returns the refactoring candidates for one maintainability property, ranked by their weight in the rating. Each candidate has an id to pass to update_finding_status.

• system_property: one of duplication, unitSize, unitComplexity, unitInterfacing, moduleCoupling, componentIndependence, or componentEntanglement.
• count: how many candidates to return. The default is 20.
• technology: filter on a technology.
• status: the statuses to include. The default is all of them: RAW, WILL_FIX, and ACCEPTED.

Security and reliability

security.get_findings returns open security findings, ranked by severity and then by exploitability, with CWE identifiers and file locations. reliability.get_findings does the same for reliability findings, such as error handling, concurrency, and resource management. Each finding has an id to pass to get_finding and update_finding_status. Both take the same parameters:

• severity_min: LOW (the default), MEDIUM, HIGH, or CRITICAL.
• model: the security model, one of ow10, sigsec, 5055sec, c25, pci4, owasvs4c, owasvs4s, or lcnc10. For reliability, sigrel or 5055rel. Leave it out for your organization’s default.
• path_prefix: only findings under this file path. Long, specific prefixes work best.
• limit: at most 100. The default is 25.
• status: the statuses to include. By default, findings marked FIXED or FALSE_POSITIVE are left out.

Open source health

opensourcehealth.get_risks returns open source dependency risks across six dimensions: vulnerability, freshness, legal, activity, stability, and management. It is the default tool for any question about open source health.

• risk_dimension: only these dimensions. A dependency matches if any of them is at or above risk_min.
• risk_min: NONE, LOW, MEDIUM (the default), HIGH, or CRITICAL.
• limit: at most 100. The default is 25.

opensourcehealth.get_vulnerabilities returns the known CVEs in your dependencies, ranked by CVSS score.

• severity_min: LOW, MEDIUM (the default), HIGH, or CRITICAL.
• limit: at most 100. The default is 25.

Findings

get_finding looks up one finding by its id, typically one you saw earlier through one of the get_findings tools.

• finding_id: the id returned with the finding.
• finding_type: security, reliability, or maintainability.
• system_property: required when finding_type is maintainability.

update_finding_status sets the status or the remark of a finding, so Sigrid reflects the agent’s decision. Open source health findings do not have a status, so this tool does not work for them.

• finding_id: the id returned with the finding.
• status: the new status, from the table below.
• remark: a note to attach to the finding.

You have to pass at least one of status and remark. The valid statuses depend on the type of finding:

Architecture

The architecture tools read Sigrid Core’s dependency graph, which describes the baseline branch as Sigrid last analyzed it. Paths have to match Sigrid’s own paths, which can differ from your local layout.

architecture.get_internal shows how the parts inside a directory relate: which parts call which, and how often.

• path: the directory. Leave it out for the system’s top-level components.

architecture.get_external_dependencies lists the direct dependencies of a file or directory, what it calls and what calls it, to find the blast radius of a change. It returns one hop per call.

• path: the file or directory. This is required.
• direction: incoming, outgoing, or all (the default).

architecture.get_worst_directories returns up to 20 directories, ranked by structure rating, worst first. The ranking is weighted by volume, so a low rating on a large component outranks the same rating on a small one.

• path: rank the directories inside this path instead of across the whole system.
• min_volume: leave out directories at or below this volume in person-years. The default is 0.2. Lower it, for example to 0.01, to see leaf directories on a small system.

Tool selection

Only the tools for the Sigrid capabilities your organization has enabled are available. If your organization has Maintainability and Security enabled, for example, you only see the tools for those.

To restrict the tools in a session further, pass the X-Enabled-Tools header with a comma-separated list of tool names:

{
  "mcpServers": {
    "axis": {
      "url": "https://sigrid-says.com/mcp",
      "headers": {
        "Authorization": "Bearer <your_sigrid_token>",
        "X-Enabled-Tools": "guardrails.quality_check, security.get_findings"
      }
    }
  }
}

Troubleshooting

Here are the problems we see most, and what solves them:

On Sigrid On-Premise, a refused connection or a 401 usually has one of three causes. Check that the URL uses your own Sigrid host, that the token was created in your on-premise Sigrid, and that your administrator has enabled the MCP server in the Helm chart.