GitLab MCP: Safe JSON Parsing in API Service

Context

GitLab’s MCP (Model Context Protocol) server includes internal tools that communicate with GitLab APIs and process JSON responses. These components must safely handle external data while remaining fully compatible with expected API payloads.

A RuboCop rule was introduced to flag unsafe JSON parsing and enforce the use of GitLab’s hardened parsing utilities.

Problem

The MCP API service was using Gitlab::Json.parse to parse HTTP responses. While functional, this method does not enforce limits on JSON size or nesting depth.

This triggered RuboCop violations and introduced potential risk when handling unexpectedly large or deeply nested JSON payloads.

My Contribution

I replaced Gitlab::Json.parse with Gitlab::Json.safe_parse in the MCP API service, aligning the implementation with GitLab’s secure parsing guidelines.

In addition, I extended the rescue clause to handle Gitlab::Json::ParserError, ensuring consistent error handling for size and depth limit violations introduced by safe_parse.

The change is intentionally minimal and preserves the existing request/response flow while adding built-in protections against malformed or excessively complex JSON input.

Example Code Change

// Before
parsed_response = Gitlab::Json.parse(response.body)

// After
parsed_response = Gitlab::Json.safe_parse(response.body)

rescue JSON::ParserError, Gitlab::Json::ParserError => e

Only the parsing mechanism and rescue clause were updated. The surrounding logic and error response structure remain unchanged.

Process Notes

The initial merge request branch eventually fell significantly behind master, which led to merge conflicts and pipeline instability.

To resolve this cleanly, I rebuilt the change on top of the latest master in a fresh branch and opened an updated merge request: !225533.

This updated MR supersedes the earlier !222565 and ensures a clean diff, accurate CI results, and minimal history noise.

Current Status

The updated merge request is based on the latest master and is currently under review.

This contribution addresses a scoped backend maintenance task and improves the robustness and security posture of the MCP server.

Reflection

This contribution demonstrates how small, targeted backend refactors can improve security and code quality without altering application behavior.

It also provided practical experience working within a large monorepo, handling branch drift, resolving CI pipeline issues, and maintaining clean contribution history in a production-scale open-source project.