Which integrations are available for this server?

Provides web search through a self-hosted SearXNG instance, returning up to 8 results with title, URL, and snippet. Optionally, pages from URLs returned in the current session can be fetched when enabled.

How do I use web-mcp?

1. Click on "Deploy Server". 2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state. 3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@web-mcp search the web for the latest yamllint release notes" That's it! The server will respond to your query, and you can continue using it as needed. Here is a step-by-step guide with screenshots.

web-mcp

by koldeda

Overview Schema Source Related Servers Discussions Admin

JavaScript

Local

web-mcp

Local web search for LM Studio / Bionic as an MCP server.

web_search(query): searches through your own SearXNG. Returns up to 8 results with title, URL and snippet.
fetch_page(url) (optional, off by default): reads the text of a page that web_search returned.

Zero npm dependencies. Node 22, built-ins only.

Security model

Layer	What it does
Container	Read-only filesystem, all Linux capabilities dropped, non-root, memory/CPU/process limits. Nothing from your Mac is mounted.
Internal network	`websearch-internal` has no route out. The server can only reach `searxng` (and `egress` in fetch mode). No internet, LAN, Tailscale or Mac.
Query	The server fixes host and path. The query only goes into `?q=`. Max 200 chars, control characters stripped.
Results	Each block is wrapped in `UNTRUSTED WEB CONTENT [id=<random>]` … `END UNTRUSTED WEB CONTENT [id=<same>]`. The id is new for every response, so a page cannot fake the end line. Marker-like text inside results is removed.
fetch_page	Accepts only URLs that web_search returned in this session (exact match). A model cannot build `evil.com/?d=<your data>`. Only http/https on ports 80/443, no credentials, max 3 redirects (each fully re-checked), 2 MB (raw and decompressed), 10 s total, text/html and text/plain only, max 5 pages / 20 attempts per session.
In-process guard	Blocks private, loopback, link-local, CGNAT/Tailscale (100.64/10), multicast, reserved and documentation ranges. Also covers IPv6, including embedded-IPv4 forms. Blocks local names (`localhost`, `.local`, `.internal`, `.home.arpa`, `.ts.net`, single-label names …).
Egress proxy	In fetch mode all traffic goes through a squid container that resolves DNS itself and refuses the same destinations. This is a second, independent implementation. It also refuses anything that isn't CONNECT to 80/443, and any client outside `websearch-internal`.

What this cannot prevent: the model putting private text into a search query. The container never holds your data, but the model does. So:

Session rule: use web search only in chats where the memory server is off and no private files or work documents are in the conversation. Keep shell access ("Allow coding") off in web sessions.

Related MCP server: SearXNG MCP Server

Files

index.js            entry point
lib/server.js       MCP protocol (newline-delimited JSON-RPC over stdio)
lib/search.js       web_search
lib/fetch.js        fetch_page
lib/http.js         HTTP client: total deadline, size caps, direct / CONNECT-proxy
lib/netguard.js     IP / hostname / URL rules
lib/text.js         markers, neutralising, HTML-to-text (linear time)
egress/squid.conf   egress proxy config
egress/verify.sh    proves sandbox + proxy behave (run on the Mac)
tools/mcp-smoke.mjs protocol conformance check for any stdio MCP server
test/               node --test suites (49 tests, local mock servers only)
Dockerfile          node:22-alpine, runs as user "node"

Setup (Mac with colima)

You need:

colima + docker running (colima status)
the searxng container from before, with json enabled under search.formats
Node 22+ for the tests (node -v)

0. Put the folder in place

unzip ~/Downloads/web-mcp.zip -d ~/.lmstudio/
cd ~/.lmstudio/web-mcp

1. Run the tests

node --test          # expect: 49 pass, 0 fail

2. Create the internal network (fixed subnet: the proxy trusts only this one)

# if you created websearch-internal earlier without a subnet, remove it first:
docker network disconnect websearch-internal searxng 2>/dev/null; docker network rm websearch-internal 2>/dev/null

docker network create --internal --subnet 172.30.99.0/24 websearch-internal
docker network connect websearch-internal searxng

If 172.30.99.0/24 collides with something on your side, pick another private /24. Change it in egress/squid.conf (acl clients src …) to match.

3. Build the image and pin the base image

docker pull node:22-alpine
docker inspect --format '{{index .RepoDigests 0}}' node:22-alpine
# -> node@sha256:abc123...   put that into the Dockerfile:
#    FROM node:22-alpine@sha256:abc123...
docker build -t web-mcp:local .

4. Check the sandbox

docker run --rm --network websearch-internal node:22-alpine sh -c '
  wget -q -T 5 -O /dev/null https://example.com && echo "BAD: internet reachable" || echo "good: no internet";
  wget -q -T 5 -O - "http://searxng:8080/search?q=test&format=json" | head -c 120; echo'

You want good: no internet followed by the start of a JSON answer.

5. Smoke test, exactly as LM Studio will start it

node tools/mcp-smoke.mjs --call web_search '{"query":"yamllint"}' -- \
  docker run -i --rm --read-only --cap-drop ALL --security-opt no-new-privileges \
  --pids-limit 64 --memory 256m --cpus 0.5 --network websearch-internal \
  -e SEARXNG_URL=http://searxng:8080 web-mcp:local

Expect ALL CHECKS PASSED and real search results in the output.

6. Register it in LM Studio (search only)

Add this next to your existing memory entry in ~/.lmstudio/mcp.json:

"web": {
  "command": "/opt/homebrew/bin/docker",
  "args": ["run", "-i", "--rm",
           "--read-only", "--cap-drop", "ALL", "--security-opt", "no-new-privileges",
           "--pids-limit", "64", "--memory", "256m", "--cpus", "0.5",
           "--network", "websearch-internal",
           "-e", "SEARXNG_URL=http://searxng:8080",
           "web-mcp:local"]
}

Then:

Restart LM Studio.
In a new chat without the memory server, enable only web.
Ask something current.

If LM Studio can't reach Docker, add "env": {"DOCKER_HOST": "unix:///Users/YOUR_USER/.colima/default/docker.sock"} to the entry.

Stop here if search is all you want.

7. Optional: enable fetch_page with the egress proxy

Start the proxy. It runs as the unprivileged proxy user (uid 13), read-only, logging to a tmpfs:

docker pull ubuntu/squid:latest
docker run -d --name egress --restart unless-stopped \
  --network bridge \
  --read-only --tmpfs /var/log/squid:uid=13,gid=13,mode=0750 --tmpfs /tmp:uid=13,gid=13 \
  --user 13:13 --cap-drop ALL --security-opt no-new-privileges \
  --memory 128m --pids-limit 64 \
  -v "$HOME/.lmstudio/web-mcp/egress/squid.conf:/etc/squid/squid.conf:ro" \
  --entrypoint /usr/sbin/squid \
  ubuntu/squid:latest -N -f /etc/squid/squid.conf
docker network connect websearch-internal egress

docker exec egress cat /var/log/squid/cache.log | tail -5   # should end with "Accepting HTTP Socket connections"

Prove it:

sh egress/verify.sh      # expect: ALL CHECKS PASSED

Switch LM Studio to fetch mode. Replace the args of the web entry with:

["run", "-i", "--rm",
 "--read-only", "--cap-drop", "ALL", "--security-opt", "no-new-privileges",
 "--pids-limit", "64", "--memory", "256m", "--cpus", "0.5",
 "--network", "websearch-internal",
 "-e", "SEARXNG_URL=http://searxng:8080",
 "-e", "WEB_ALLOW_FETCH=1",
 "-e", "EGRESS_PROXY=http://egress:3128",
 "web-mcp:local"]

Restart LM Studio. fetch_page now shows up next to web_search.

Logs

What	Where
Queries and fetched URLs	The server's stderr, one line each. LM Studio shows it in its MCP/developer log.
Proxy decisions	`docker exec egress cat /var/log/squid/access.log`. `TCP_DENIED/403` means refused, `TCP_TUNNEL/200` means allowed.

Troubleshooting

"cannot reach SearXNG":
- docker ps: is searxng running?
- Is it connected to websearch-internal?
- After a reboot, run colima start (or brew services start colima to start it automatically).
"SearXNG did not return JSON": add json under search.formats in ~/searxng/config/settings.yml, then run docker restart searxng.
egress container exits immediately: docker logs egress. If it complains about permissions, first try without --read-only, then send the log for review. Don't just drop all the hardening.
Every fetch returns "blocked by egress proxy … 503": squid can't resolve names. Check with docker exec egress cat /var/log/squid/cache.log. Adding --dns 9.9.9.9 (or another resolver you trust) to the egress docker run fixes it.
A page returns almost nothing: it's JavaScript-rendered, or behind a bot check such as "Client Challenge". fetch_page reads plain HTML only; there is no browser.
IPv6: if squid's cache.log says IPv6 has not been enabled, squid ignores its IPv6 rules. It then also cannot make IPv6 connections, so that is safe.

Check any MCP server

tools/mcp-smoke.mjs works for any stdio MCP server, not just this one. Run it before trusting a server in LM Studio:

node tools/mcp-smoke.mjs -- node ~/.lmstudio/memory/mcp-server/index.js

This server cannot be deployed

Give agents eyes on the webHosted screenshot MCP. Send a URL, get a pixel-perfect image. $0.002 per grab. Works with Claude, Cursor, and Codex.grabbit.liveAd

Maintenance

ActivityMaintained

ResponsivenessNo issues

Resources

GitHub Repository

Need Help?

Related Servers

Related MCP Connectors

search-mcp B
dev.ioconnex
Web search and clean-text page fetch for AI agents, with SSRF protection.
Search Browser Automation Web Scraping
21
Not Human Search A
ai.nothumansearch
Search the agentic web. 4,100+ sites, 11 tools incl. check_url + verify_mcp for probe-before-use.
Search Browser Automation Web Scraping
143
seltz-ai-seltz-mcpOAuth
ai.seltz
Provides AI assistants with access to Seltz's powerful Web Search capabilities.
Search Browser Automation Web Scraping
2
foundrynet-search B
io.github.FoundryNet
x402-gated web search gateway. Tools: search, search_enriched.
Search RAG Systems
2

Related MCP Servers

SearXNG MCP Server
Search Browser Automation
ngc-shj
A
license
A
quality
D
maintenance
Enables web search, image search, and news search through a self-hosted SearXNG instance. Provides privacy-focused meta-search capabilities aggregating results from multiple search engines.
Updated 9 months ago (2026-01-05 06:13 UTC)
3
752 npm
1
MIT
SearXNG MCP Server
Search Browser Automation
franktheglock
A
license
A
quality
D
maintenance
Enables web search capabilities via a SearXNG instance, allowing for filtered searches across categories like news, images, and science. It supports multi-language queries, time-range filtering, and safe search options to provide formatted results with metadata.
Updated 10 months ago (2025-12-02 19:05 UTC)
1
752 npm
MIT
SearxNG MCP Server
Search Browser Automation
IceWreck
F
license
A
quality
D
maintenance
Enables privacy-respecting web, image, video, and news searches through SearxNG metasearch engine, with options for URL fetching and time filtering.
Updated 7 months ago (2026-03-07 18:54 UTC)
5
-
Web Explorer MCP
Browser Automation Web Scraping Search
l0kifs
A
license
A
quality
D
maintenance
Enables private web search and webpage content extraction using a local SearxNG instance, prioritizing user privacy and autonomy.
Updated a year ago (2025-10-25 14:27 UTC)
2
2
MIT

web-mcp

Security model

Files

Setup (Mac with colima)

0. Put the folder in place

1. Run the tests

2. Create the internal network (fixed subnet: the proxy trusts only this one)

3. Build the image and pin the base image

4. Check the sandbox

5. Smoke test, exactly as LM Studio will start it

6. Register it in LM Studio (search only)

7. Optional: enable fetch_page with the egress proxy

Logs

Troubleshooting

Check any MCP server

Maintenance

Resources

Related MCP Connectors

Related MCP Servers

SearXNG MCP Server

SearXNG MCP Server

SearxNG MCP Server

Web Explorer MCP