MCP server that provides legal research tools for searching court opinions, dockets, and court information from the Court Listener API, integrated with Microsoft Teams bot and Azure OpenAI
Court Listener MCP server has 4 well-named, read-only tools with documented input schemas and descriptions. Tool names follow verb_noun convention (search_opinions, get_opinion_details, search_dockets, get_court_info). All parameters have type declarations and descriptions. However, output schemas are not formally documented in the code, responses are assembled as anonymous JSON objects without a declared schema structure. Parameter descriptions are adequate but generic in places (e.g., 'Search query text' could specify expected format or examples). No error handling guidance visible, errors would return raw API responses without recovery hints. Tool descriptions are moderate length (50-90 chars) but lack context about when to use each tool vs. alternatives. Missing parameter constraints (e.g., jurisdiction enum values are documented but not enforced as formal constraints). No tool annotations (readOnlyHint) present, though all tools are read-only. Date parameter handling shows manual workarounds for framework bugs, indicating potential usability friction. Responses strip some API metadata but inconsistently, plain_text is truncated to 2000 chars, suggesting awareness of context limits, but no pagination or limit documentation for search results.
Get information about courts, including court names, jurisdictions, and metadata
Get detailed information about a specific opinion by ID
Search for dockets/cases by case name, docket number, court, and other criteria
Search for court opinions by keywords, court, date range, and other criteria
Output schemas not formally documented. Responses are anonymous JSON objects assembled at runtime without a declared schema contract. LLMs cannot plan downstream tool calls or extract structured fields reliably.
No error handling guidance or recovery hints. Raw API errors would return to LLM without context about whether to retry, ask user, or abandon. Pattern requires actionable error messages.
Parameter constraints not formally declared. 'jurisdiction' field documents valid values as text string 'F, FD, FB, ...' instead of enum constraint. LLMs cannot validate inputs and may hallucinate invalid values.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 64 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 54 | - | v1 |
No pagination documented or enforced for search results. Code limits results to 10 items (.Take(10)) but does not return count, next_cursor, or explain pagination to LLM. Search results may silently truncate.
Tool annotations missing. All 4 tools are read-only but toolAnnotations=false. MCP 2026-07-28 spec recommends readOnlyHint annotation to signal safe tools for agent optimization.
Date parameter handling requires manual workaround for framework bug (see code comment re: GH Issue). Users/LLMs must pass dates as YYYY-MM-DD strings; framework does not parse DateTime naturally. Undocumented friction point.