Loading ToolBench
A Platform MCP prototype for Filevine legal matter management. Lets MCP clients search matters, read documents/tasks/notes, file well-scoped tasks, and pull cross-matter health reports. Covers LOIS Workflows, DataBridge exports, and Platform MCP surfaces. Read-only by default with optional write access via FILEVINE_MCP_ALLOW_WRITES=1. Every call is audited. Includes mock mode with bundled fixture data.
Server has 6 tools with complete input schemas and descriptions. Naming follows verb_noun convention (matter_health_report, get_deadlines, run_workflow, create_task, add_note, export_snapshot). Descriptions are substantive (100-250 chars), explaining WHAT and WHEN. However, parameter descriptions lack detail on constraints (e.g., priority enum values, date format specifics), and output schemas are not formally documented. Error handling is present but generic. No tool annotations (readOnlyHint/destructiveHint) despite clear read/write distinction. Audit trail and permission gating are implemented server-side, which is correct, but not surfaced in tool metadata.
Add a note to a matter. Requires FILEVINE_MCP_ALLOW_WRITES=1 to execute. Returns the created note with noteId, projectId, authorId, date, and text.
Create a new task on a matter. Requires FILEVINE_MCP_ALLOW_WRITES=1 to execute. Returns the created task with taskId, projectId, title, assignee, dueDate, status, and priority.
Export a versioned, schema-stable CSV snapshot of the organization's matters, documents, tasks, and notes. Returns a manifest with row counts and a frozen column contract. Implements DataBridge-style point-in-time extraction.
Returns all upcoming deadlines across matters, including derived dates from deadline-chain engine. Surfaces dates that matter_health_report cannot see, such as discovery meet-and-confer windows and governmental notice dates.
Portfolio-level triage: returns a summary of all matters with status, activity, and risk indicators. Surfaces what needs attention across the entire matter portfolio.
Output schemas not documented. Tools return structured data (task objects, note objects, CSV manifests) but LLMs cannot see field contracts. Downstream tool chaining and field extraction require guesswork.
Parameter constraints not specified. 'priority' in create_task lacks enum (critical|high|normal|low). 'due_date' format stated as ISO but no validation guidance. 'workflow_id' not constrained to workflows.json keys.
Tool annotations missing. create_task, add_note, run_workflow are destructive/write operations but lack destructiveHint. matter_health_report and get_deadlines are read-only but lack readOnlyHint. Agents cannot infer safety without explicit metadata.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
Execute or preview a declarative workflow from workflows.json. Defaults to dry_run=true to show the plan before execution. Workflows can create tasks, add notes, and perform other actions on matters. Write access requires FILEVINE_MCP_ALLOW_WRITES=1.
Error handling generic. Code references FilevineError but tool responses do not show structured error classification (retryable vs user-fixable vs fatal). LLMs receive raw exceptions without recovery guidance.
Permission model not exposed in tool metadata. FILEVINE_MCP_ALLOW_WRITES environment variable gates writes, but tools do not declare required scopes (read:matters, write:tasks, write:notes). Agents cannot self-configure least-privilege access.