Tools that help your AI answer real questions.

A small, opinionated directory of free and open MCP servers. Connect one to your AI of choice — Claude, ChatGPT, Cursor, anything that speaks the protocol — and ask in plain English. The server fetches real, current information from a trusted source and hands it back so the AI can answer without making things up.

Maintained by David Loor · MIT licensed · No accounts, no tracking

· · ·

§1The premise

Large language models are good at language and unreliable at facts. They hallucinate names, invent citations, and confidently mis-quote numbers — especially when the answer requires looking something up that wasn't on the open web during training.

An MCP server is a small, single-purpose program that knows how to fetch one kind of real information well. The AI calls it the way you'd call a library function. The answer the AI gives you is then grounded in whatever the server returned, not in whatever the model half-remembers.

Protocol: Model Context Protocol, an open standard from Anthropic.
Transport: Streamable HTTP. Each server is just a URL.
Cost: Free. You already pay your AI app; these add nothing.
Privacy: No storage. Each request is handled in memory and discarded.

· · ·

§2The servers

No. 01

find-trials

live

Searches ClinicalTrials.gov for studies matching a condition, location, and recruitment status — and returns the contact phone numbers a real person can call.

Asked like

My dad was just diagnosed with pancreatic cancer. Are there any trials recruiting in Texas?

The AI returns matching trials with what each one is studying, where it's based, the inclusion criteria, and phone numbers you can call to ask about enrolling.

https://mcp.davidloor.com/find-trials

Source · ClinicalTrials.gov Repository Readme Privacy

No. 02

jobscope-mcp

live

Searches open roles across ATS platforms — Greenhouse, Lever, Ashby — in one pass. No logins, no cookies, no recruiter spam. Just a ranked list of postings that match what you asked for.

Asked like

I'm a senior backend engineer open to remote roles. What's hiring right now?

The AI fans out across dozens of company career boards and returns a ranked list of matching roles with direct apply links you can open right away.

https://mcp.davidloor.com/jobscope-mcp

Source · Greenhouse, Lever, Ashby Repository Readme

Planned · expected within the year

drug-info Look up drug labels and adverse-event data from DailyMed and the FDA.

rare-disease Navigate the rare-disease landscape with Orphanet as the primary source.

· · ·

§3Connecting it

Pick the AI you use. Each one has a slightly different name for the same idea — a place to add a custom server URL.

Claude claude.ai or desktop app

Open Settings → Connectors.
Click Add custom connector.
Paste the server URL, e.g. https://mcp.davidloor.com/find-trials.
Save. Claude tests the connection and lists the tools it found.
Open a new chat and ask your question in plain English.

ChatGPT plus, team, or enterprise

Open Settings → Connectors. Depending on your plan it may live under Beta features or Developer.
Add a server. Paste the URL.
Confirm. ChatGPT will connect and list the available tools.

Cursor, Codex, other clients config-file based

Most clients accept a streamable-HTTP MCP server via a config panel or a config file.
In Cursor, it's Settings → MCP. Add a server with the URL above.
In OpenAI Codex CLI, edit ~/.codex/config.toml and add an mcp_servers entry pointing at the URL.

· · ·

§4Questions people ask

Is it really free?: Yes. No account, no subscription, no usage cap. You pay your AI app — Claude, ChatGPT — for the chat itself; these helpers cost nothing extra.
Do you save my questions or anything about me?: No. Each request lives in memory for as long as it takes to answer, then it's gone. There is no log, no analytics, no profile. Full privacy commitment.
Should I rely on these for important decisions?: These are information lookups, not advice. For medical, legal, or financial decisions, always confirm with a qualified human before acting. Treat the result as a fast, honest hand-off of real data — useful, but not a verdict.
Can I run my own copy?: Yes. Everything is open source and MIT-licensed. Each repository carries its own self-hosting instructions and deployment notes.

· · ·

§5Privacy

These servers store no user data. None. Not in logs, not in analytics, not for “quality.”

Each question is handled in memory, used to compose one response, and then discarded. There are no accounts to create, no histories to browse, no profiles being built. Each repository publishes its own PRIVACY.md with the full commitment in plain English.