MCP server for accessing N8N node information through Claude Desktop
This server has basic tool definitions with schemas and descriptions, but significant gaps in parameter documentation, output schema clarity, and error handling guidance. All 7 tools are explicitly registered with names and descriptions in index.js. However, 5 of 7 tools have zero required parameters (empty input schemas), limiting LLM agency. Parameter descriptions are minimal or absent. Output schemas are not documented, responses return unstructured text rather than typed objects. No error handling guidance, no enum constraints, and no actionable error messages. The server lacks the structured, LLM-optimized design needed for production agents.
Check GitHub API rate limit status
Get the first N lines of a node's source code
Get details about a specific N8N node
List all AI/LangChain N8N nodes
List all N8N nodes from both core and AI packages
List community N8N nodes from npm
Search for N8N nodes containing a keyword in both core and AI packages
Five of seven tools have no input parameters and empty schemas. This severely limits LLM reasoning and composability. Tools like list_all_nodes and list_community_nodes cannot accept filters, pagination, or limits, forcing agents to process full result sets or make multiple discovery calls.
Output schemas are not documented. All tool responses return unstructured text (via { type: 'text', text: ... }). LLMs cannot parse structured fields or chain outputs to subsequent tool calls. Responses should declare typed fields (e.g., 'nodes: [{name, package, category}]') so downstream tools can reliably extract IDs and references.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 27 | - | v1 |
Parameter descriptions are missing or trivial. 'node_name' in get_node_details is described only as 'Name of the N8N node', no guidance on format, case sensitivity, or how to discover valid names. 'keyword' in search_nodes lacks any description of matching behavior (substring? regex? case-insensitive?). Per the rubric baseline, 100% of A+ tools document all parameters; this server has ~50% underdocumented params.
No enum constraints or parameter validation guidance. The 'lines' parameter in get_node_code_snippet defaults to 50 but has no min/max bounds. If an LLM passes lines=10000, will it timeout or return a huge response? Enums and ranges should be declared explicitly in schemas to guide LLM input.
No pagination support. list_all_nodes and list_ai_nodes return unbounded results as plain text ('Found 1200 nodes... and 50 more'). Per the rubric baseline, tools returning lists must accept limit and offset/page parameters and return a total count. This prevents agents from efficiently discovering large node catalogs.
Error handling is minimal and non-actionable. The code catches GitHub API errors silently ('// LangChain package might not exist') with no guidance to the LLM. If a node is not found or the API rate limit is hit, the LLM receives no error classification, recovery suggestion, or actionable message. Per the rubric, error responses must tell the LLM 'what to do next.'
Description quality is inconsistent and brief. 'List all N8N nodes from both core and AI packages' (62 chars) is borderline; 'Check GitHub API rate limit status' (36 chars) falls below the 50-char baseline for effective LLM selection. The rubric baseline average is 194 chars for A-tier tools; these descriptions are 30-60% of that. Missing context on WHEN to call each tool and WHAT to do with results.
No security or permission declarations. Tools query GitHub API but do not document what scopes they require (read:public_repos, etc.). No audit logging of tool calls. If an LLM somehow malforms a GitHub URL or injects a path traversal payload, there is no sanitization.
GitHub token is stored in process.env.GITHUB_TOKEN but no validation or rotation guidance is provided. If the token leaks or expires, the server fails silently. No circuit breaker, retry backoff, or rate-limit awareness beyond check_rate_limit (which is a manual inspection tool, not automatic enforcement).