Loading ToolBench
Editor-grade TypeScript, Markdown, and JSON intelligence for coding agents.
Type Atlas provides 15 well-structured TypeScript tools with complete JSON schemas and consistent descriptions. All tools are READ_ONLY and properly namespaced. Naming follows verb_noun convention (hover, completions, diagnostics). Descriptions are present and substantive (avg ~120 chars), explaining WHAT each tool does and WHEN to use it. Schemas include proper type definitions, enums (e.g., includeDiagnostics: off|summary|full), and parameter constraints (limit: 1-100). However, several gaps prevent a higher score: (1) No output schemas documented, LLMs cannot predict return structure; (2) Error handling is implicit, not explicit, no recovery guidance in descriptions; (3) Some parameter descriptions lack format/constraint details (e.g., 'workspace' and 'file' paths lack validation rules); (4) No idempotency or retry guidance despite tools being read-only; (5) module_exports has complex nested parameters (path, surface, query) with minimal guidance on interdependencies.
Return TypeScript's source-wide missing-import fixes as a Codex patch.
Discover quick fixes and refactors for a range, or resolve a displayed action to a Codex patch. Stable source-wide actions have dedicated tools. The MCP does not modify files.
Return a bounded completion page at a source position. Set resolve to include upstream details for that page or raw to return every unresolved candidate.
Report diagnostics for the TypeScript projects you have touched — the compiler's own whole-program check, run once per project.
Return the symbol hierarchy of a document: declarations, their nesting, and their locations.
Return TypeScript's source-wide fix-all action as a Codex patch when applicable.
No output schemas documented. LLMs cannot predict return structure, field names, or types. Downstream tool chaining is impossible without trial-and-error.
Error handling is implicit. Descriptions do not explain failure modes, recovery steps, or when to retry. E.g., 'workspace not found' or 'file parse error', no guidance on what the LLM should do next.
Path parameters (workspace, file) lack validation rules. Descriptions do not specify: absolute vs relative, required format, or what happens if path is invalid. LLMs may pass malformed paths.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Return a Codex patch with TypeScript formatting applied to a document.
Return type and documentation hover at a position.
Inspect exported members of a module with optional type filtering and pagination.
Return the TypeScript organize-imports action as a Codex patch.
Return TypeScript's source-wide unused-code removal as a Codex patch.
Return a Codex patch that moves files and updates affected references. The MCP does not modify files.
Return a Codex patch for a TypeScript-project symbol rename at a source position. The MCP does not modify files.
Return nested selection ranges at one or more source positions.
Return overload and parameter information at a call site.
module_exports has complex interdependent parameters (path, surface, type, query) with minimal guidance on valid combinations. E.g., when is 'type' required? What does 'surface: all' do if 'path' is specified?
No pagination guidance for tools returning lists (completions, module_exports, diagnostics). Descriptions mention limit/offset but do not explain total count, next_cursor, or when results are truncated.