Loading ToolBench
Monorepo code indexing for AI agents with AGENTS.md support. Provides bounded context discovery, full-text search, domain mapping, and agent instruction retrieval across single and multi-repo setups.
This MCP server demonstrates solid definition quality with clear naming conventions, comprehensive descriptions, and well-structured schemas. All 5 tools follow verb_noun naming (get_*, search_), have detailed descriptions (194-356 chars, well above the 34-char p10 baseline), and include input parameters with type definitions and descriptions. The server is domain-focused on code repository indexing with AGENTS.md support, which is a narrow, well-understood scope. However, output schemas are not explicitly documented in the visible code, responses are string-based rather than structured JSON objects. Error handling provides context (available contexts listed, repository not found messages) but lacks actionable recovery paths in some cases. Security is well-handled (all tools are READ_ONLY, no secrets exposed). Tool composition is clean with no overlapping responsibilities.
Load the full AGENTS.md file for a specific bounded context. AGENTS.md files contain domain rules, terminology, constraints, and agent instructions. Always read the AGENTS.md before modifying code in a bounded context.
Get the full domain map showing all bounded contexts, their AGENTS.md files, scope, and relationships. Use this to orient yourself in the codebase before diving into code.
Retrieve a file's content along with its bounded context metadata. Use this when you need the full file rather than search result chunks.
Search the indexed codebase for code relevant to a query. Returns trimmed snippets inline and writes full results to a file. Use `get_file` or read the full results path to see complete code chunks. Supports fuzzy matching, compound identifier splitting (PascalCase, snake_case, kebab-case). AGENTS.md references are included as path annotations — use `get_agents_context` to load full content. Requires deep index - run `smooth-code-repo-index index` if not available.
Resolve a file path to its governing bounded context and AGENTS.md file. Use this to identify which domain you're working in before making changes.
Output schemas not documented. Tools return unstructured strings (e.g., `public string GetAgentsContext(...)` returns markdown text) rather than structured JSON objects. LLMs cannot reliably parse multi-line markdown responses or plan downstream tool chains without knowing what fields to extract.
Error responses lack recovery guidance. When context not found or repo ambiguous, the error message lists alternatives (e.g., 'Available contexts: ...') but does not explicitly suggest next steps like 'Try get_domain_map to see all available contexts.' This increases agent confusion.
search_codebase result pagination not explicitly shown. Description mentions 'trimmed snippets inline and writes full results to a file' but does not document the limit parameter's default (10) or max value, or how LLMs should handle file-based results vs inline results. Ambiguous multi-format output.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | A | 80 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 50 | - | v1 |
No tool for discovering available repositories in multi-repo mode. If an agent is deployed against multiple repos and the config changes, there is no canonical tool to query 'list_repos()' or 'get_active_repos()'. Agents must hard-code or infer repo aliases.