This server helps you to understand how ASTs are structured by OXC parser. Also useful to quick check for your code which should be valid syntax.
The OXC AST MCP server has 3 READ_ONLY tools with visible input schemas and descriptions. However, there are consistent quality gaps: tool names lack action verbs (docs, parse, check are not ideal verb forms); descriptions are present but lack critical details about when/why to use each tool; parameter descriptions are generic and omit constraints; output schemas are not documented; and error handling is absent. The server does not demonstrate the composition patterns needed for a production-grade tool. STDIO-only transport further limits this server's production utility.
Accepts a code snippet and extension. Extension should be one of `js`, `mjs`, `cjs`, `jsx`, `ts`, `mts`, `cts`, `tsx`. Returns both syntactic and semantic errors by default, but can also disable semantic checks if `check_semantic` is set to `false`. You can optionally specify parser options to customize parsing behavior.
Accepts a regex string to filter only matched structs and enums.
Accepts a code snippet and extension. Extension should be one of `js`, `mjs`, `cjs`, `jsx`, `ts`, `mts`, `cts`, `tsx`. You can optionally specify parser options to customize parsing behavior. You can use `docs` tool to see the documentation for each AST node.
Tool names lack action verbs. 'docs', 'parse', and 'check' are imperative forms but do not follow the canonical verb_noun pattern. Per arcade.dev/patterns/tool, LLMs rely on names like 'get_docs', 'parse_code', 'check_syntax' to infer intent before reading descriptions. Current names are ambiguous: 'docs' could mean 'view docs', 'generate docs', 'list docs'; 'check' could mean 'run checks', 'validate', 'lint'.
Output schemas are not documented. The tool descriptions state what inputs are accepted but do not explain what fields the LLM will receive in the response. For 'parse', does it return an AST object with 'type', 'children', 'errors' fields? For 'docs', does it return a list of struct/enum definitions or a single matched definition? Without response schemas, agents cannot plan downstream calls or extract the right fields (rubric DIMENSION 1.D critical check: 'Document the output schema'). Per mxe:strip-api-responses and mxe:include-chaining-ids, responses must be predictable and include references needed for chaining.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 40 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 48 | - | v1 |
Parameter descriptions are generic and omit constraints. Example: 'query' in docs is described as 'Optional regex string to filter only matched structs and enums', does not specify regex flavor, performance limits, or required format. 'ext' in parse/check lists valid values ('js, mjs, cjs, jsx, ts, mts, cts, tsx') inline in the description instead of as an enum constraint. Additionally, descriptions like 'Whether to parse regular expressions or not (default: true)' are minimal, no guidance on when to set to false or performance implications.
No error handling or recovery guidance. The source code does not show error responses, validation of inputs, or actionable error messages. If an invalid regex is passed to 'docs', or invalid code to 'parse', the LLM receives no guidance on what went wrong or how to fix it.
Tool descriptions lack context for selection. 'docs' is described as 'Accepts a regex string to filter only matched structs and enums', but does not answer: When should the LLM call this vs. parse or check? Is it a discovery tool? What is the response format? 'parse' says 'You can use `docs` tool to see the documentation for each AST node', but this embeds a tool recommendation in a description rather than letting the LLM infer it from composition. State WHAT the tool does, WHEN to use it, and any prerequisites.' Current descriptions are WHAT-only.
Missing tool annotations. The server does not declare readOnlyHint, destructiveHint, or idempotentHint in tool definitions. All three tools are READ_ONLY (per the provided metadata), but this is not communicated via schema annotations.
Parameter type definitions incomplete. Input schema for 'docs' has 'query' as type 'string', which is correct, but no minLength, maxLength, pattern, or enum. Input schema for 'parse' has 'ext' listed as a string with description containing the enum values inline, the JSON Schema should use an enum array. Input schemas lack type constraints (e.g., are parse_regular_expression, allow_return_outside_function booleans or strings?). Validate inputs early and return clear error messages.'