Loading ToolBench
MCPB Desktop Extension for Indian tax drafting. Provides tools for drafting Indian Direct-Tax appellate documents (CIT(A), ITAT, DRP, High Court, PCIT, CIT) with privacy firewall, multi-agent orchestration, and verification workflows.
Server has only 2 tools with minimal schema definition. list_case_types has empty input schema ({}), which is acceptable for a read-only discovery tool. get_case_type_format has a single string parameter with description, but no output schema is documented anywhere in the source. Descriptions are present but generic (under 100 chars each). No error handling guidance, no parameter validation rules, no output structure documentation. Tool names follow verb_noun pattern (list_, get_) which is good, but the server lacks the depth and rigor expected of production tools. The FULL_ORCHESTRATION_SCRIPT in the code references 6 additional tools (create_case_folder, save_artifact, save_draft_as_docx, get_agent_instructions, get_pleading_base) that are NOT registered as MCP tools, they are only mentioned in documentation. This is a critical gap: the orchestration script assumes tools that do not exist in the MCP interface.
Get the drafting format SKILL.md for a given case type. Returns the full SKILL.md content for the case type, including statutory anchors, format placeholders, and the drafting checklist. Also appends format-from-user.md if present in the same skill folder.
List all case types this connector can draft. Returns a dict with the list of case types, a short description of each, AND an acronym disambiguation map (Indian advocacy acronyms / Form references / statutory section references -> case_type). CRITICAL: the user's acronym IS the source of truth for case_type classification. Do not infer case_type from input-document content.
Missing output schema documentation. Neither tool documents what fields/structure the response contains. LLMs cannot plan downstream calls without knowing the response shape.
Orchestration script references 6 tools (create_case_folder, save_artifact, save_draft_as_docx, get_agent_instructions, get_pleading_base, and one more) that are NOT registered as MCP tools. The script is documentation only; agents cannot invoke these tools via MCP.
No error handling guidance. Descriptions do not explain what happens if case_type is invalid, or what errors the tools return. LLMs have no recovery path.
get_case_type_format parameter 'case_type' description says 'One of the slugs returned by list_case_types()' but does not enumerate valid values or provide an enum constraint. LLMs may hallucinate invalid case_type values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Tool descriptions are generic and under 100 characters. They lack context on WHEN to use each tool, what the user intent is, and how they fit into the drafting workflow. Descriptions should be 50-200 chars and LLM-optimized.