#059Technical4 min

Examine MCP servers and their configuration

Read an MCP configuration before installation and understand what a static component check can establish.

Hackaiz team · updated 21/11/2026

Illustration: a founder builds an AI app.
Illustration: a founder builds an AI app. Illustrative photo, not a screenshot.

01Read configuration as a request for access

MCP connects an AI assistant to tools or data sources. A server configuration can therefore grant authority that is not obvious from a friendly component name. Review the launch command, arguments, accessible folders and external destinations before enabling it. A tool intended to read one project should not silently receive access to your whole disk. This article concerns reviewing submitted component files, not granting a server permission to run.

Hackaiz looks for risky component patterns such as remote scripts piped into a shell, secret-file reads, hidden instructions, encoded commands and configurations that expose the whole disk. It also considers dangerous combinations: reading secrets and sending data outside can be more serious together than either capability appears in isolation. Treat these as review signals backed by file evidence, not a conclusion based only on the server’s name or popularity.

02Use Check before installing

Sign in and open Check before installing. You can paste the MCP configuration into the text area or select Or choose files to upload the relevant configuration, such as mcp.json. The Component name field helps identify what you are reviewing. If files are selected, the pasted-text field is disabled, so verify that the chosen files are really the version you intend to install. Select Check to launch the review.

The page excludes secret files such as .env files and keys, and shows excluded paths as not checked. Remove real credentials from configuration text before sharing it; do not turn a review of permissions into a disclosure of the credentials themselves. A configuration may refer to a package whose full source was not submitted. Reading that configuration does not establish the behavior of every file the package might later download or execute.

  1. 1Inspect the proposed command, arguments, folder access and destinations.
  2. 2Paste the configuration or choose its files on Check before installing.
  3. 3Read any excluded-file notice before submitting.
  4. 4Select Check, then review the evidence and any suggested safer alternative.

03Interpret the component verdict and coverage

The component guide distinguishes Safe, Attention, Danger and Incomplete. These component verdicts are different from the severity groups in an application assessment. Safe means the performed checks did not identify the listed risky behavior, not that the server is certified trustworthy. Incomplete means part of the content could not be read or checked; it must not be treated as approval to install.

Read the exact flagged line and the reason it matters. A command that retrieves a remote script and pipes it to a shell deserves scrutiny even if the surrounding instructions look routine. Access to environment secrets combined with an unfamiliar network destination calls for understanding what leaves the machine. Review safer alternatives where offered, but do not assume every finding has an automatically applicable repair. The final installation decision stays with the person responsible for the environment.

04Keep static limits and repeat-check costs visible

This static component review reads submitted files without executing the server. Behavior that appears only at runtime can be missed. File contents are analyzed in memory and not retained; only the verdict summary is saved in history. Every new check uses credits, including a repeat with identical content. If the page reports a failure, review history and balance before trying again rather than assuming the previous attempt produced a complete verdict.

The same component check is available through the command line as hackaiz check <path>; it can return an error on Danger and export a machine-readable security report using --sarif. Choose the interface that fits your review process without confusing automation with complete coverage. If you instead need to assess a deployed website or API used by an agent, that is a separate authorized analysis requiring domain ownership proof through _hackaiz DNS TXT, the HTTPS verification file, or the project AI file-creation instruction. A component upload is not such a proof.

Frequently asked questions

Was this article helpful?

Related articles