A Model Context Protocol (MCP) tool for analyzing software licenses using Scancode data.
The server defines 5 tools with schemas and descriptions, but quality is inconsistent. Tool names lack action verbs and are unnecessarily prefixed with 'mcp_ScancodeMCP_' (redundant branding). Descriptions are present but generic and lack context about when to use each tool or what happens next. Parameters have type annotations but several lack meaningful descriptions, e.g., 'random_string' in two tools is a dummy parameter masking poor API design (no-parameter tools should use empty inputSchema, not fake params). Schema validation is minimal; no enums, ranges, or constraints are declared. Error handling is absent, no guidance on recovery, retries, or failure modes. No output schema is documented. The server depends on a static JSON file (license_analysis_detailed.json) with no validation of its contents or graceful degradation if missing. Overall: definitions exist but violate multiple high-value patterns around naming, parameter clarity, and error guidance.
Clause-by-clause legal analysis of all licenses detected in a file, including obligations, risks, and compatibility.
Legal compatibility verdict and explanation for two license types (e.g., MIT vs GPLv3).
Clause-by-clause legal summary of a license (obligations, risks, compatibility, etc).
Lists all files with copyleft, unknown, or commercial-unfriendly licenses, with a legal warning for each.
Lists all files with high-risk/problematic licenses and provides a legal risk summary for each license type.
Tool names prefixed with 'mcp_ScancodeMCP_' which is redundant branding that violates naming clarity. LLMs will see the full name (e.g., 'mcp_ScancodeMCP_analyze_license_file') and conflate the prefix with the action, increasing cognitive load. Standard pattern is verb_noun (e.g., 'analyze_license_file').
Two tools ('summarize_license_risks', 'list_high_risk_files') declare a dummy 'random_string' parameter to work around the absence of support for no-parameter tools. This is an anti-pattern: it confuses LLMs into thinking they must provide a string, adds noise, and violates the single-responsibility principle. Solution: use empty inputSchema (per MCP spec) or document that these tools require no input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 35 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 34 | - | v1 |
No input validation or constraint declarations. Parameters lack enums, ranges, regex patterns, or allowed-value lists. For example, 'linesToRead' (number) has no min/max, 'licenseName' (string) has no enum of valid licenses. This forces LLMs to guess valid inputs and leads to runtime failures without actionable error messages.
No output schema documentation. All tools return { content: [{ type: 'text', text: '...' }] } but the structure, fields, and format of the 'text' value are not documented. LLMs cannot plan downstream steps or extract structured data. This violates the pattern for documented return types.
No error handling or recovery guidance. The code checks if 'licenseData' is loaded but returns a generic 'License data not loaded...' message. No guidance on what to do next, whether to retry, or how to diagnose the root cause. Per the pattern, errors should categorize as retryable/user-fixable/fatal and suggest next steps.
Hardcoded Windows path ('C:\\Users\\Admin\\Desktop\\LICENSE_MANAGER\\') as default parameter value. This reduces portability and will fail silently on Linux/Mac systems. Should accept an environment variable, config file, or require the LLM to pass the path explicitly.
Description includes example values in parentheses (e.g., 'licenseName' description: 'The license name to summarize (e.g., MIT, GPL-3.0, unknown)'). Per the rubric, LLMs tend to reuse example values literally; use enums or format constraints instead.
No pagination support. 'summarize_license_risks' and 'list_high_risk_files' iterate all high-risk items and return them in a single free-text response. For large datasets, this will exceed token limits and degrade LLM reasoning. Should support limit/offset and return structured results.