Pod

Available as Markdown and JSON. Pod is also available over MCP.

Reported issues for wet-mcp

Pod holds 13 of 13 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.

Back to wet-mcp.

Most discussed

SearXNG server fails with uvx

Describe the bug

When using the recommended command, uvx --python 3.13 wet-mcp (docs), the SearXNG server doesn't start successfully and web searches are unsuccessful.

To Reproduce

Steps to reproduce the behavior:

  1. Start the MCP server:
    uvx --python 3.13 wet-mcp
  2. Send an initialize command: (paste this into the same terminal and hit Enter)

Read the thread · 2026-04-06 · closed · external user · 4 comments

OAuth: valid Bearer rejected at /mcp + headless-client (Hermes) connect fails

Context

Kết nối Hermes Agent (self-hosted MCP client, NousResearch) tới wet-mcp đã deploy qua OAuth.

Observed

  1. Hermes auto-discovery fail vì mcp-core 401 thiếu resource_metadata (xem issue mcp-core).
  2. Ép --auth oauth: PKCE flow ra auth URL hợp lệ; user authenticate; /token trả Bearer hợp lệ (resource=http://oci-wet-mcp:8080/mcp, expires 1h).
  3. NHƯNG đưa Bearer đó tới /mcp vẫn 401 Unauthorized. Nghi audience/resource validation mismatch (token issued cho internal resource…

Read the thread · 2026-05-28 · closed · 3 comments

[Epic] Phase 2 (v1.x+1.y) — Context7-level docs search

Tracking issue for v1.x+1.y release. Context7-level library documentation search: version-aware, project-pinned, token-efficient.

Features

Docs search — Context7 parity

Grouped under search tool (not separate docs tool) for lean surface. Actions:

Read the thread · 2026-04-20 · closed · 3 comments

[Epic] Phase 1 (v1.x.y) — web-core migration + search polish + media slim

Tracking issue for v1.x.y release. Migrate extract pipeline to n24q02m-web-core shared primitives + search polish + media slim + hygiene cleanup.

Blocker: web-core LLM fix

Phase 1 is blocked on web-core fix to remove hardcoded gemini-2.5-flash default in web_core.scraper.selector_inference. Tracked separately at n24q02m/web-core — see linked issue.

Web Extract — migrate to web-core ScrapingAgent

Replace direct crawl4ai usage in sources/crawler.py with…

Read the thread · 2026-04-20 · closed · 3 comments

feat: upgrade to qwen3-embed v2.0.0 — tiny local models + offline reranking

Summary

Upstream qwen3-embed is releasing v2.0.0 with two major changes that affect this project:

  1. Smaller embedding model: tiny-embed-v1 (~60MB ONNX INT8) replaces Qwen3-Embedding-0.6B (573MB) — ~10x smaller, faster inference, multilingual 50+ languages + code support
  2. New local reranking: tiny-reranker-v1 (~70-90MB ONNX INT8) — cross-encoder reranker that runs entirely locally via ONNX Runtime

Impact on wet-mcp

Embedding (existing)

Read the thread · 2026-04-12 · closed · 3 comments

fix: cf container index fails with ModuleNotFoundError: No module named 'qwen3_embed'

Read live from prod D1 wet-docs, 2026-08-03.

Evidence

library             dv ver      status     cc chunks   idx_at state    err
fastapi             27 latest   indexed    50     50 1785732468 running
fastapi:python       0 latest   pending     0      0     null failed   ModuleNotFoundError: No module named 'qwen3_embed'
httpx                0 latest   pending     0      0     null running
httpx:python         0 latest   pending     0      0     null None

versions.index_error…

Read the thread · 2026-08-03 · closed · 2 comments

[Epic] Phase 3 (v2.0.0) — extract agent + interact, media analyze removal (BREAKING)

Tracking issue for v2.0.0 major release. Higher-level extract actions (agent-driven multi-step research, interactive browser ops) + media analyze action full removal. Auto-migrate on startup for schema changes (seamless for users).

Features

extract(action="agent") — higher-level research

Firecrawl /agent parity. User prompt → wet performs:

  1. search(web, query=<derived_from_prompt>) → top N URLs.
  2. ScrapingAgent.scrape each URL in parallel (bounded concurrency). 3.…

Read the thread · 2026-04-20 · closed · 2 comments

feat: Zero-env-config setup via Relay Page (ECDH + PIN)

Problem

Currently, users may need to configure multiple env vars: SEARXNG_URL, LITELLM_PROXY_URL, API_KEYS (for SDK mode), embedding backend settings, etc. The 3-mode configuration (local/SDK/proxy) adds complexity.

This is an industry-wide MCP problem: 88% of MCP servers require credentials, 79% pass them via plain env vars.

Proposal

Add a Relay Page — a web UI for zero-env-config setup with end-to-end…

Read the thread · 2026-03-24 · closed · 2 comments

Most recent

fix: search() kills CF container -- D1 neighbour prefetch binds 300 params against a 100 cap

Symptom

Calling search(action="docs", library="fastapi") against the deployed Cloudflare worker kills the container. The MCP client only ever sees:

Error executing tool search: Server disconnected without sending a response.

Reproduced 4/4 times.

Evidence

Captured live from wrangler tail wet-mcp-worker (2026-08-03):

1785733634022 ok         ContainerProxy  http://vectorize.internal/query
1785733634204 exception  ContainerProxy  http://d1.internal/query   <== FAILS…

[Read the thread](https://github.com/n24q02m/wet/issues/1627) · 2026-08-03 · closed · 0 comments

### fix: cf-d1 get_best_version ignores the status filter and the fallback that DocsDB applies

Found during the #1624 column parity pass, deliberately left out of PR #1625 to keep a prod-blocking fix narrow.

## Divergence

`DocsDB.get_best_version` (db.py):

```sql
-- exact target
SELECT * FROM versions WHERE library_id = ? AND version = ? AND status = 'indexed'
-- then, if that misses, fall back
SELECT * FROM versions WHERE library_id = ? AND status = 'indexed'
 ORDER BY indexed_at DESC LIMIT 1

db_cf.get_best_version filters no status on either branch, and does not fall back at…

Read the thread · 2026-08-03 · closed · 0 comments

fix: multi-user embed resolver ignores DISABLE_LOCAL_EMBED, so CF indexing dies on ModuleNotFoundError

Observed on the live CF deployment right after v3.7.0-beta.2 shipped the index-state observability columns (#1610).

Symptom

search(action="docs", library="fastapi") against https://wet.n24q02m.com/mcp leaves doc_chunks empty. D1 now records why:

index_state = failed
index_error = ModuleNotFoundError: No module named 'qwen3_embed'

Stable across 15 polls over 8 minutes, so not a transient.

Root cause

embedder.py:524 resolve_embed_backend_for_request() has exactly…

Read the thread · 2026-08-03 · closed · 1 comment

Cannot listen 8080 port under Docker envrionment when PUBLIC_URL is not specified

https://github.com/n24q02m/wet-mcp/blob/88db575ae16d041ce070e0f4c0479f322b5584b0/src/wet_mcp/server.py#L3047-L3061

For users who use their own app in docker containers, they have to deploy wet-mcp as a HTTP service just for accessing it from other containers, and use this service with single-user mode. Under this scene the PUBLIC_URL is not needed, but the server wasn't bound to a fixed 8080 port, and was only bound to 127.0.0.1, so the server will find a random port, and noone can…

Read the thread · 2026-08-02 · closed · external user · 1 comment

[Phase 3] Unified MCP Core migration — work in progress

Status

Phase 3 "Unified MCP Core" migration is in active development. Contributors: please coordinate before opening PRs against main until Phase J lands.

What changes for wet-mcp

Read the thread · 2026-04-11 · closed · 1 comment

The remaining reports are on the project's issue tracker.