Loading ToolBench
Git-native reasoning layer for tracking architectural decisions, plans, systems, and constraints. Provides MCP tools for context retrieval, diff review, validation, and AI agent workflows.
Intent MCP provides 13 well-named, read-only tools for architectural decision management. Tool names follow verb_noun convention (intent_context, intent_list_plans, intent_search). All tools have descriptions (avg ~120 chars, within baseline 34-392 range). Input schemas are present with typed properties and descriptions. However, output schemas are NOT documented, critical gap for LLM planning. Parameter descriptions are present but sparse (e.g., 'Working directory' lacks format guidance). No error handling guidance visible. No tool annotations (readOnlyHint/destructiveHint). Two write tools (intent_rebuild_links, intent_export) lack confirmation/dry-run patterns. Missing pagination/limit guidance for list tools despite baseline expectation.
Get intent context relevant to a task. Scores plans/decisions/systems/constraints by keyword match against the task description, pulls in cross-referenced items transitively. Falls back to full context when nothing matches.
Review the current diff against intent and validate all files. Returns a checklist and flags whether unlinked changes suggest a new decision record is needed.
Get all intent context (plans, decisions, systems, constraints) for the project. Optionally filter by system.
Export intent context as a formatted block, or write it to a target file.
List all constraints in the project
List all architectural decisions in the project
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., plan IDs, decision references). Baseline: 100% of A+ tools document return types.
List tools (intent_list_*) lack pagination parameters (limit, offset, page_size) and result count guidance. Baseline: tools returning lists should accept pagination and return total count. Risk: large result sets blow context window.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
List all plans in the project
List all system definitions in the project
Rebuild the .intent/links/index.json index
Get intent context relevant to the current git diff. Use before committing to check what reasoning applies.
Full-text search across all intent files (plans, decisions, systems, constraints). Returns matches ranked by relevance with excerpts.
Get a health overview of the .intent/ directory: plan counts, active/draft plans, links index state, and validation summary.
Validate all .intent/ files and check referential integrity. Returns issues grouped by file.
Write tools (intent_rebuild_links, intent_export) lack confirmation/dry-run pattern. Agents cannot preview changes before committing. No error recovery guidance if operation fails.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe read-only tools from destructive ones without explicit hints. Current spec (2026-07-28) recommends annotations.
Parameter descriptions lack format/constraint guidance. E.g., 'cwd' says 'Working directory' but does not specify absolute vs relative, or validation rules. Baseline: 100% of A+ tool params have actionable descriptions.