- Reliability 8/20
- The README claims 29 tests runnable via npm test, but no CI workflow evidence is present in the repository materials, and those tests themselves require a live, debug-enabled TradingView Desktop instance to run — they cannot be independently verified from static files. More notably, the documentation directly contradicts itself: one heading says 'Tool Reference (78 MCP tools)' while the Architecture section says 'MCP over stdio (84 tools)' — a mismatch indicating docs are out of sync with the actual tool count. The tool also operates via undocumented internal Electron/CDP interfaces, and the README itself warns these 'can change or break without notice in any TradingView update,' which is a self-disclosed reliability risk. Per the static calibration rule (unverifiable execution evidence caps this dimension below 12), combined with the internal inconsistency and uncontrolled external dependency, this is scored 8.
- Security and permissions 11/20
- The tool communicates only with a locally running TradingView Desktop instance over localhost:9222, stores no credentials, and makes no external network calls — genuine positives. However, several operations have real, potentially irreversible side effects with no described confirmation step: pine_save overwrites scripts saved to the TradingView cloud, alert_delete/draw_clear remove user data, and ui_evaluate offers general-purpose arbitrary JS execution inside the Electron renderer — a broad privilege surface with no documented scoping or confirmation gating. The README also candidly discloses that the tool's core function (programmatic chart data access) may violate TradingView's Terms of Use, risking account bans — an honest disclosure but not a technical mitigation. None of this rises to the malware/credential-theft/covert-exfiltration red-line bar, so no red-line is applied, but the clear gaps in least-privilege and confirmation handling justify a score of 11.
- Maintenance 9/20
- The project has a high star count (5,681) suggesting attention, but 199 open issues with no static evidence of triage speed or maintainer responsiveness. More critically, the repo metadata reports License: NOASSERTION while the README claims an MIT license backed by a LICENSE file — this contradiction suggests GitHub could not detect a properly formatted license text, a governance red flag. There is also no version number, changelog, or security-contact channel described anywhere in the README. Combined, the license inconsistency and missing governance signals yield a score of 9.
- Documentation 14/20
- Overall documentation is fairly comprehensive: it covers installation, per-platform launch scripts, a CLI-to-MCP-tool mapping table, a full categorized tool reference, a context-management strategy section, and a dedicated disclaimer/usage-limits section — above-average coverage. Deductions apply for the internal tool-count inconsistency (78 vs 84), the lack of parameter-level specs for individual tools (only one-line descriptions are given, not types/required fields/return shapes), and the fact that running the test suite requires a live TradingView connection without further troubleshooting guidance for that prerequisite. Usable but with notable detail gaps, so it is scored 14.
- Setup experience 11/20
- The setup path is reasonably clear: clone, npm install, launch TradingView with a debug flag (via provided Mac/Windows/Linux scripts or manually), add the MCP config, and verify with tv_health_check — plus a table of common install locations across platforms, which is a genuine plus. However, per the static calibration rule, setup cannot exceed 15 without verifiable, committed CI plus tests covering the key setup paths; here the test suite itself depends on a live, debug-enabled TradingView Desktop instance and there is no CI evidence, so it doesn't qualify as verifiable execution proof. The process also requires manually editing config file paths and deliberately opening a debug port that is disabled by default, against internal interfaces the authors admit can break at any time — a multi-step, somewhat fragile flow, warranting a score of 11.