Oathe Security Badge

oathe / scan / reports / safe

Is shreeyachand/goodreads-mcp safe?

Yes. shreeyachand/goodreads-mcp is safe to install. Oathe's behavioral security audit gave the goodreads-mcp skill by shreeyachand a trust score of 89/100 with 6 findings, none critical or high. Report updated

https://github.com/shreeyachand/goodreads-mcp

89
SAFE

Is shreeyachand/goodreads-mcp safe to install?

goodreads-mcp is a well-structured, read-only MCP server that wraps Goodreads' unofficial web surfaces (JSON autocomplete, Next.js Apollo state, AppSync GraphQL, RSS feeds). The code contains no credential-harvesting logic, no install hooks, no suspicious network destinations, and no SKILL.md prompt injection. The primary security concern is indirect prompt injection: all tools return user-generated Goodreads content (reviews, descriptions, shelf names) with only HTML stripping applied, so a malicious Goodreads account could plant adversarial instructions in fetched text. Additionally, the runtime JS-bundle parsing to discover the AppSync endpoint reduces long-term auditability. These are design-level concerns inherent to any tool returning third-party user content, not evidence of malicious intent.

What security issues were found in shreeyachand/goodreads-mcp?

Category Scores

Prompt Injection 91/100 · 30%
Data Exfiltration 90/100 · 25%
Code Execution 82/100 · 20%
Clone Behavior 95/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 74/100 · 5%

Findings (6)

MEDIUM Indirect prompt injection via Goodreads user-generated content -9 ▶

All tools that return review text (get_reviews), book descriptions (get_book, search_books), shelf content (get_shelf), or list titles (book_lists) pass user-generated strings through only HTML stripping and entity unescaping. Any Goodreads user can craft a review or book description containing adversarial instructions ('Ignore all previous instructions...') that will be delivered verbatim to the host LLM. The risk escalates when this skill is co-installed with tools that have write or execute capabilities.

Evidence: _clean_text() in server.py strips HTML tags and unescapes entities but applies no prompt-injection filtering. get_reviews returns rev.get('text') after _clean_text; get_book returns _clean_text(book.get('description')).

LOW Runtime remote-configuration loading from Goodreads JS bundles -18 ▶

graphql_config() fetches the Goodreads _app JavaScript bundle at runtime using regex to extract the AppSync API key and endpoint. While no code is executed from the bundle, the skill's effective network behavior and authentication target are determined by whatever Goodreads serves at that URL. If Goodreads were compromised or if a future bundle embeds a malicious endpoint, the skill would transparently redirect GraphQL traffic.

Evidence: client.py: APP_CHUNK_RE, APPSYNC_PAIR_RE, parse_appsync_config(), parse_appsync_endpoint() — endpoint and key resolved from https://www.goodreads.com/giveaway and /_next/static/chunks/pages/_app-*.js at import time.

LOW Skill reads ~/.config/goodreads-mcp/config.json from user home directory -10 ▶

config.py resolves the user's Goodreads profile ID from an optional config file at ~/.config/goodreads-mcp/config.json. While this file is skill-specific and benign, the pattern of reading from the user's home directory could be a concern in environments where that path could be manipulated or overlap with sensitive material.

Evidence: config.py: CONFIG_PATH = Path.home() / '.config' / 'goodreads-mcp' / 'config.json'

INFO compare_books fans out to up to 10 simultaneous page fetches -26 ▶

compare_books iterates over up to 10 book IDs, calling get_book() for each. Each get_book() fetches a Goodreads page plus Apollo state. This amplifies exposure to injected content and also sends up to 10 sequential HTTP requests to Goodreads per tool call, which could be rate-limited or used to profile the user's queries.

Evidence: server.py: _MAX_COMPARE = 10; compare_books() calls get_book(bid) in a for loop.

INFO Network connection to GitHub CDN only during install 0 ▶

The git clone made a single HTTPS connection to 4.163.8.144:443 (GitHub CDN). No other unexpected outbound connections were observed during installation. This is completely expected behavior.

Evidence: auditd: SOCKADDR saddr=020001BB04A308900000000000000000 (4.163.8.144:443) from git-remote-http

INFO Canary file reads attributable to audit framework, not skill 0 ▶

Filesystem monitoring detected reads of .env, .ssh/id_rsa, .aws/credentials, .npmrc, .docker/config.json, and GCP credentials at both 13:39:35 (pre-clone, session 4) and 13:39:57 (post-analysis). The pre-clone reads are from the monitoring setup (sudo, session 4). The post-analysis reads align with the audit framework's post-install canary integrity scan. No skill code reads these paths. Canary integrity report confirms all files intact.

Evidence: inotify: OPEN /home/oc-exec/.env at 13:39:35 (pre-clone) and auditd PATH records at timestamp 1794663597 (post-analysis). Canary report: ✅ All canary files intact.

Should I install shreeyachand/goodreads-mcp?

Oathe's verdict for shreeyachand/goodreads-mcp is SAFE with a trust score of 89/100. Recommendation: Install with caution.