MCP server providing WebDAV tools for file operations over HTTP/WebDAV transport
webdav-mcp provides 8 tools with complete JSON schemas and descriptions visible in src/main.zig. All tools have clear verb-noun naming (list, read, write, delete, mkdir, move, copy, stat) and reasonable descriptions (105-240 chars). However, several quality issues reduce the score: (1) Parameter descriptions lack detail on constraints and formats, e.g., 'path' params have no guidance on how to format paths, whether leading slashes are required, or what happens with relative paths. (2) No output schemas documented, the code returns text results but LLMs cannot predict field structure without explicit documentation. (3) Error handling is minimal, code catches errors but provides no guidance on recovery, retryability, or what the LLM should do next. (4) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk patterns. (5) Schemas are present but sparse, no examples, no default values documented, no enum constraints. The tools are functional but would benefit from richer parameter documentation and explicit output schemas.
Copy a file or directory on WebDAV.
Delete a file or directory on WebDAV.
List files and directories at a WebDAV path. Returns name, size, type, last-modified, ETag, and content-type for each entry.
Create a directory on WebDAV.
Move or rename a file or directory on WebDAV.
Read the contents of a file from WebDAV. Returns the file content as text.
Get metadata for a single file or directory on WebDAV (PROPFIND Depth:0). Returns type (file/dir/missing), size, last-modified, ETag, and content-type. Use this to check existence or get file info without listing a whole directory.
Output schemas are not documented. Code shows tool results are returned as plain text wrapped in ContentBlock structures (type=text), but LLMs cannot predict response field structure or plan downstream tool calls without explicit schema documentation. Pattern baseline: 100% of A+ tools have documented return types.
Parameter descriptions lack actionable constraint details. 'path' parameter is described only as 'Directory path relative to WebDAV root (e.g. '/' or '/projects/')', no guidance on: leading slash requirement, maximum length, forbidden characters, what happens with relative paths, or URL encoding. LLMs need explicit constraints to avoid malformed input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
Write content to a file on WebDAV. Creates or overwrites the file.
No tool annotations present despite clear risk patterns. Tools like delete (DESTRUCTIVE), write (WRITE), and mkdir (WRITE) should carry readOnlyHint/destructiveHint/idempotentHint annotations to help LLMs understand safety boundaries and avoid unintended irreversible actions.
Error handling provides no recovery guidance. Code in writeError() sends error codes and messages but does not categorize errors as retryable, user-fixable, or fatal. LLMs receive raw errors (e.g., network timeouts, auth failures, not-found) with no actionable next steps.
No confirmation or dry-run support for destructive operations. delete and move with overwrite=true can cause data loss with no opportunity for the LLM to preview or confirm. Irreversible operations should support a dry-run or explicit confirmation step.
Parameter enums not used for constrained inputs. For example, 'recursive' is a boolean with unclear semantics, should it default to false? What happens if the server rejects infinite depth? Similar: 'overwrite' param on move/copy lacks an enum or clear statement of mutual exclusivity.
No pagination or result limiting documented. list tool with recursive=true could return thousands of entries, causing context window exhaustion. No documented cap on result count, no next_cursor or offset/limit parameters, no guidance on chunking large directory listings.
read tool max_bytes default of 1 MB is large for text streaming. No guidance on what happens with binary files, encoding issues, or partial reads. If a 1 MB file is text with invalid UTF-8, the tool may return unparseable gibberish.