n8n Forum MCP

An MCP server that searches the n8n community forum so AI agents (Claude, etc.) can use it as a knowledge source — pairs nicely with the official n8n Docs MCP.

Solved threads on the forum (those with an accepted answer) are effectively vetted community documentation: real problems with confirmed fixes. This server makes them searchable and readable from your agent.

Read-only and anonymous — it only hits the public Discourse search/topic endpoints. No login, no credentials.

Tools

• search_forum(query, solved_only=False, max_results=8) — search the forum. Set solved_only=True to return only threads with an accepted answer (vetted solutions). Returns each thread's solved-status, title, category, link, and excerpt — plus, for solved threads, a snippet of the accepted answer inline so the vetted fix shows up in the search itself.

• get_thread(topic_id) — read a thread by id (from search_forum). Surfaces the accepted answer first, then the posts.

Adaptive recall

Discourse search ANDs every term, so pasting a whole error message or a full sentence ("OpenAI node messages must be a non-empty array got null despite having inputs") tends to match nothing — while the same intent as a few key terms ("messages non-empty array") finds the solved threads. search_forum handles this for you: it runs the exact query first (precision), then progressively broadens to the most distinctive 4 → 3 → 2 terms only if results are thin (recall), merging solved threads to the top. So you can hand it a raw error string and still get hits.

Related MCP server: Discourse MCP

Install

git clone https://github.com/acunningham-ship-it/n8n-forum-mcp
cd n8n-forum-mcp
python3 -m venv .venv
./.venv/bin/pip install -r requirements.txt

Use with Claude Code

claude mcp add n8n-forum -- /absolute/path/to/n8n-forum-mcp/.venv/bin/python /absolute/path/to/n8n-forum-mcp/forum_mcp.py

For Claude Desktop / other clients, add an equivalent stdio server entry pointing at .venv/bin/python forum_mcp.py.

CLI (no MCP client needed)

forumsearch.py works standalone for quick lookups:

python3 forumsearch.py search "airtable rate limit"   # all results
python3 forumsearch.py solved "azure openai api version"  # accepted-answer threads only
python3 forumsearch.py thread 300911                  # read a thread + its accepted answer

License

MIT