wet
AGENTS.md19 starsChanged 9 months agoApache-2.0View on GitHub
Python MCP Server: web search, content extraction, library docs, structured extraction. Xem AGENTS.md va README.md de hieu architecture va configuration. Script entry points: wet (wet.cli:main) and wet-mcp (legacy alias). wet search and other server-backed subcommands talk to the running server over streamable HTTP; with no server running they exit 2 and print a hint to run wet server start. wet docs reindex|import|reembed work without a server.
- Reads credentials
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 60
Default timeout: 30 seconds per test
- Distinctive: line 117
No container image and no Cloudflare deploy are published.
- Very specific: line 124
Commit message: enforce feat/fix prefix
- Very specific: line 138
All suites spawn the real HTTP server (python -m wet.server) on an ephemeral loopback port against a tmp instance home.
≈ “Common” is an estimate: at least 1 in 5 of the 36,368 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md - wet
Python MCP Server: web search, content extraction, library docs, structured extraction.
Xem AGENTS.md va README.md de hieu architecture va configuration.
Cau truc
src/wet/-- Package chinh (src layout)server.py-- FastMCP server (orchestrator, file lon nhat); always the HTTP MCP endpoint (no stdio mode)cli.py--wetcontrol plane:server start|stop|status,config,token hash,users path,docs reindex|import|reembed,search,warmupruntime.py-- Bridge to hull-core (settings, auth mode, users store, per-sub namespace)config.py-- Pydantic Settings (singleton)cache.py,db.py,embedder.py,reranker.py-- Infrastructureidentity.py-- Coherent browsing identity across the strategy chaindocs_import.py,docs_reembed.py-- Offline docs maintenance (no running server needed)setup.py,setup_tool.py-- First-run SearXNG install + warmup logic (MCP-callable)sources/-- Data source integrations (crawler, docs, searxng, search backends/strategies)alembic/-- Docs DB migrations
tests/-- Mirror source modules
Lenh thuong dung
uv sync --group dev # Cai dependencies
uv build # Build package (hatchling)
uv run ruff check . # Lint
uv run ruff format --check . # Kiem tra format
uv run ruff check --fix . && uv run ruff format . # Fix
uv run ty check # Type check (ty lenient config)
uv run pytest # Test tat ca (integration excluded by default)
uv run pytest -m integration # Chi integration tests
uv run pytest tests/test_config.py::test_function_name -v # Test don le
uv run wet server start # Chay server (background; --foreground de giu terminal)
uv run wet server status # pid / lock / HTTP liveness
uv run wet server stop
# Mise shortcuts
mise run setup # Full dev env setup
mise run lint # ruff check + ruff format --check + ty check
mise run test # pytest
mise run fix # ruff check --fix --unsafe-fixes + ruff format
mise run dev # uv run wet server start --foregroundScript entry points: wet (wet.cli:main) and wet-mcp (legacy alias).
wet search and other server-backed subcommands talk to the running server over
streamable HTTP; with no server running they exit 2 and print a hint to run
wet server start. wet docs reindex|import|reembed work without a server.
Cau hinh quan trong
- Python 3.13 bat buoc -- 3.14+ KHONG tuong thich do SearXNG
requires-python = "==3.13.*"trong pyproject.toml- Ruff: line-length 88, target py313, rules E/F/W/I/UP/B/C4, ignore E501
- ty: lenient (unresolved-import, unresolved-attribute, possibly-missing-attribute all "ignore")
Pytest
asyncio_mode = "auto"-- KHONG can@pytest.mark.asyncio- Default timeout: 30 seconds per test
- Integration/live/full tests excluded by default
addopts = "-m 'not integration and not live and not full'"trong pyproject.toml
Server, bind & auth
- One transport: the HTTP MCP server (
wet server start, orpython -m wet.server). - Bind:
[server] host/portin~/.wet/config.toml;WET_HOST/WET_PORTenv override (container deployments). - Per-port lifecycle lock (
LifecycleLock("wet", port), lock files under~/.config/mcp/locks) -- one instance per port, several ports can run side by side. [server] authmodes:no-auth-- loopback bind only; the server refuses to start on a non-loopback host.token-- single bearer token; store only its hash (wet token hash <token>).multi-- per-user tokens from[server] users_file(default~/.wet/users.toml, seewet users path); data is namespaced per sub under~/.wet/subs/<sub>/.
- Client config points at the HTTP endpoint (see the manual example below).
Env vars
- KHONG co prefix ung dung (day la open-source MCP server)
- LLM/Embed/Rerank/Chat: per-task provider cells
[models.embed|rerank|chat|jev_score]in~/.wet/config.toml(eachbase_url+api_key+model, OpenAI-spec). OpenRouter is the pre-wired default for every cell.- Credentials:
api_keyin the cell, orHULL_<TASK>_API_KEYenv injection (HULL_EMBED_API_KEY/HULL_RERANK_API_KEY/HULL_CHAT_API_KEY/HULL_JEV_SCORE_API_KEY— env wins over the file). OPENROUTER_API_KEYenv serves any cell that keeps the OpenRouter defaultbase_url— one key can power every task.- Empty embed/rerank cell -> local ONNX (fastretrieval); unkeyed chat cell -> LLM features off.
- Credentials:
- Local model cache:
FASTRETRIEVAL_CACHE_PATH. - Search backends (
SEARCH_BACKENDSCSV, runtime fallback — thử từng backend, rơi xuống backend kế khi error/empty):searxng(self-host) |tavily|brave|exa|kagi(keyed,KAGI_API_KEY) |openrouter(OPENROUTER_API_KEY,openrouter:web_searchserver tool) |firecrawl(key optional — không key vẫn thử keyless,FIRECRAWL_API_KEY) |duckduckgo|startpage(credential-free HTML scrape, uvx-safe; datacenter IP có thể bị bot-challenge → chain tự advance). - SearXNG:
WET_AUTO_SEARXNG(default true),SEARXNG_URL(external mode) - Browser render backends (headless leg of
extract):BROWSER_BACKENDS(CSV escalation chain:native|browserless|invisible; empty =native),BROWSERLESS_URL/BROWSERLESS_TOKEN(self-host browserless),STEALTHFOX_BINARY(path to a patched Firefox build for theinvisibletier — skips the engine download, seal still verified),CAPSOLVER_API_KEY(optional key-gated captcha tier) - Identity layer (one coherent person across the chain):
WET_IDENTITY_SEED(int pin; default 0 = derive once and persist per namespace under~/.wet/subs/<sub>/identity.json),IDENTITY_PROFILE_DIR(persistent browser profile dir; default~/.wet/subs/<sub>/profiles). Requires the optionalhull-core[identity]extra (invisible-core); without it strategies keep legacy behavior (warn once). First identity build may download a GeoIP mmdb (egress) and resolve locale/timezone from the egress IP —timezone_mismatchbehind a proxy falls back to en-US/UTC, never raises to MCP callers - Robots policy:
RESPECT_ROBOTS_TXT(defaultfalse); settrueto enforcerobots.txtfor the ScrapingAgent extract chain and Crawl4AIcrawl,sitemap, andlist_mediaactions - Disable-local toggles (skip heavy in-process fallbacks per capability):
DISABLE_LOCAL_BROWSER,DISABLE_LOCAL_SEARCH,DISABLE_LOCAL_EMBED,DISABLE_LOCAL_RERANK - Docs DB backup is an external concern (e.g. rclone of
~/.wet/); the server has no built-in sync. - Secrets: skret SSM namespace
/wet-mcp/prod(regionap-southeast-1)
Manual config example
wet server start # binds [server] host/port from ~/.wet/config.toml{
"mcpServers": {
"wet": {
"type": "http",
"url": "http://127.0.0.1:<port>/mcp"
}
}
}With auth = "token" or "multi", add the client's bearer token header
("headers": {"Authorization": "Bearer <token>"}).
Release & Deploy
- Conventional Commits. Tag format:
v{version} - CD: workflow_dispatch, chon beta/stable
- Pipeline (
cd.yml):benchmark-contractgate ->release(PSR v10) ->publish-pypi(uv publish) ->publish-mcp-registry->sync-marketplace - No container image and no Cloudflare deploy are published.
Pre-commit hooks
- Ruff lint (
--fix --target-version=py313) + format - ty type check
- pytest (
--tb=short -q --timeout=30) - Commit message: enforce
feat/fixprefix
Luu y quan trong
- Lazy imports ben trong functions cho heavy deps va tranh circular deps
- MCP tools return error strings (
return "Error: ...") -- KHONG raise exceptions - Graceful fallback chains: Cloud -> Local, Tier 0 -> 1 -> 2 -> 3
match action:cho tool action dispatchasyncio.to_thread()cho wrapping sync operations- Embedding luu tai 768 dims (default). Doi embedding MODEL = doi vector space -> B2 identity guard CHAN boot (set REINDEX_ON_MODEL_CHANGE=true de re-embed). 768-dim chung chi giu table khong vo schema; KHONG cho mix vector tu 2 model khac nhau (cung dims van rac search).
- Renovate: Python upgrades DISABLED
Protocol & live tests
All suites spawn the real HTTP server (python -m wet.server) on an ephemeral
loopback port against a tmp instance home; there is no external E2E driver.
tests/test_protocol_gate.py-- protocol gate over the production HTTP transport, runs from the source tree (default suite).tests/test_live_protocol.py---m live(real server + stub SearXNG).tests/test_full_live.py---m full(real upstream services; keys from skret/wet-mcp/prod).tests/test_live_mcp.py-- standalone live script (defines no pytest tests; run it directly).
uv run pytest tests/test_protocol_gate.py
uv run pytest -m live
uv run pytest -m full
uv run python tests/test_live_mcp.pyReferences: ~/.claude/skills/mcp-dev/references/protocol-test-coverage.md, ~/.claude/skills/mcp-dev/references/secrets-skret.md (per-server credential layout), ~/.claude/skills/mcp-dev/references/multi-user-pattern.md (per-sub isolation).
Other versions1 project uses a version of this.
Word for word · 1
- n24q02m/wet-mcp★ 19100% same
Edited · 0
None found.
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 28,693 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository7 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- compareSkill
- fact-checkSkill
- lock-project-stackSkill
- research-topicSkill
- scrape-batchSkill
- wetSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

