What is the Model Context Protocol (MCP)?

MCP is an open standard developed by Anthropic in 2024 that lets AI assistants connect to external tools and data sources through a common interface. An MCP server exposes a set of named tools with JSON Schema input contracts. An MCP-capable client — Claude Desktop, Cursor, ChatGPT custom connectors, and others — discovers those tools and lets the model call them as part of a conversation. AdmitBase runs a public MCP server so AI assistants can query admissions data directly, without scraping or relying on stale training data.

Do I need an API key to use AdmitBase MCP?

No, not for the three public tools. search_schools, get_school_stats, and calculate_match_score all work anonymously — just point your MCP client at https://admitbase.com/api/mcp. Anonymous traffic is rate-limited to 60 requests per minute per IP. The two personalised tools — save_my_stats and compare_to_applicants — require a free API key tied to an AdmitBase account, generated at https://admitbase.com/profile#api-keys and sent as an Authorization: Bearer header.

Which AI clients work with AdmitBase MCP?

Any client that supports the Streamable HTTP transport added to the MCP spec in 2025. Confirmed working: Claude Desktop (via the mcp-remote bridge), Claude.ai remote connectors, Cursor, Windsurf, Continue.dev, ChatGPT custom connectors. New MCP clients ship constantly, and any agent framework that speaks MCP (LangChain MCP adapter, OpenAI Agents SDK MCP support, etc.) can integrate with one line of config.

What data can I query through AdmitBase MCP?

Admissions statistics for over 500 professional schools across seven programs: law (ABA 509 data for 252 US schools plus international), medical (100 US schools, 17 Canadian schools), dental, MBA, pharmacy, veterinary, and optometry. For each school you can pull ranking, acceptance rate, 25th/50th/75th percentile GPA and standardised test scores, class size, tuition, employment outcomes, and certification pass rates. The match-score tool combines a user's GPA and test score with school data to return a Safety/Target/Reach/Far-Reach classification plus admission probability.

How fresh is the data?

We refresh from official sources annually as they publish. Law school numbers come from ABA 509 disclosures (updated December each year for the prior cycle). Medical school data comes from AAMC FACTS and per-school admissions reports. Dental from ADEA, pharmacy from AACP, veterinary from AAVMC, optometry from ASCO. Each school record exposes a data_year field so AI assistants can cite the vintage of the numbers they return.

Is the MCP server free?

Yes. The public endpoint is free for AI assistants, agents, and developers. AdmitBase makes money from premium consumer subscriptions on admitbase.com (calculators, analytics, multi-school ROI, comparison tools), not from the MCP server. Heavy automated use beyond the rate limits — 60 requests per minute per IP anonymous, 600 per minute authenticated — is the only constraint.

AdmitBase MCP: Use AdmitBase Inside Claude, Cursor, and ChatGPT

What is an MCP server?

The Model Context Protocol is an open standard published by Anthropic in late 2024. It defines a single, JSON-RPC-based wire format that AI assistants can use to discover and invoke external tools. Before MCP, every AI client wired up its own bespoke plugin system, and every tool author had to ship one integration per assistant. MCP collapses that mess: build the tool once, expose it over MCP, and it works with Claude Desktop, Claude.ai connectors, Cursor, Windsurf, Continue, ChatGPT custom connectors, and the growing list of MCP-aware agent frameworks.

An MCP server is a small program that publishes a list of tools — each with a name, description, and JSON Schema input contract — and handles the actual work when an assistant asks to call one. Servers can run locally over stdio (the original transport, useful for sandboxed access to your filesystem) or remotely over Streamable HTTP (the transport the spec settled on for hosted services). AdmitBase runs in the remote configuration.

What AdmitBase MCP exposes

The server publishes five tools. Three are public; two require authentication.

Public tools

search_schools — Find professional schools by program (law, medical, dental, MBA, pharmacy, veterinary, optometry), name pattern, or ranking range. Returns ranking, acceptance rate, GPA/test percentiles, class size, and a link back to the AdmitBase school page.
get_school_stats — Pull a full statistics record for one school by slug or name match: 25th/50th/75th percentiles, tuition (resident and non-resident), employment rate, certification pass rate, location, and data vintage.
calculate_match_score — Given a user's GPA and test score, return either a Safety/Target/Reach/Far-Reach categorisation for a single school or a ranked list of the best fits across an entire program. The math is the same model used on admitbase.com: percentile-based using each school's 25/50/75 distribution, weighted per program (60 percent test + 40 percent GPA for law, 50/50 for medical and dental, weighted differently for veterinary and MBA), with an acceptance-rate adjustment for ultra-selective schools and a splitter-friendliness correction.

Authenticated tools

save_my_stats — Persist an applicant's GPA, test score, state, and notes under their API key. Useful for multi-turn agent workflows that need to remember the user's profile between sessions.
compare_to_applicants — Compare a user's saved stats against AdmitBase's anonymous community outcome reports (accepted, waitlisted, rejected). Returns percentile rankings versus actual reporters plus a GPA-bucketed acceptance-rate breakdown.

Anonymous use — paste the URL

If you only need the public tools, integration is a one-liner. Point any MCP-capable client at:

https://admitbase.com/api/mcp

For Claude Desktop, this means adding an entry to claude_desktop_config.json using the mcp-remote bridge package (Claude Desktop currently speaks stdio natively and uses mcp-remote to wrap any HTTP MCP server):

{
  "mcpServers": {
    "admitbase": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://admitbase.com/api/mcp"]
    }
  }
}

Cursor and Claude.ai connectors accept the URL directly with no bridge needed. ChatGPT custom connectors take the URL through Settings → Connectors → Add custom connector. Each client refreshes its tool list on connection; the five AdmitBase tools appear immediately.

Personalised use — generate an API key

For the two authenticated tools, generate a free API key in your AdmitBase profile (https://admitbase.com/profile#api-keys). Keys start with amb_ and are shown once at creation. Pass them as Authorization: Bearer amb_... on every MCP request. In Claude Desktop with mcp-remote:

{
  "mcpServers": {
    "admitbase": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://admitbase.com/api/mcp",
        "--header",
        "Authorization: Bearer amb_YOUR_KEY_HERE"
      ]
    }
  }
}

The same Bearer token unlocks the personalised tools while leaving public tool behaviour unchanged. Keys can be revoked at any time from the profile page — revocation takes effect immediately on the next request.

Example prompts

Once your client is connected, the assistant should automatically invoke AdmitBase tools when the conversation calls for admissions data. Useful prompts:

"Use AdmitBase to find the top 20 medical schools and tell me which of them my 3.7 GPA and 515 MCAT would be a Target for."
"Pull NYU Law's 25th, 50th, and 75th percentile LSAT and GPA numbers from AdmitBase, plus their most recent acceptance rate."
"Compare a 3.85 GPA / 168 LSAT applicant against all T-14 law schools using AdmitBase's match calculator. Group by Safety / Target / Reach."
"What is the cheapest in-state tuition among the top 50 ranked dental schools? Use AdmitBase."

Rate limits and abuse handling

Anonymous traffic is rate-limited to 60 requests per minute per IP. Authenticated traffic gets 600 per minute per key. The server returns JSON-RPC error code -32029 with a retry_after_seconds hint when limits are exceeded — your client should back off and retry. All requests are logged to a query log table with IP hashes (never raw IPs) for analytics and abuse detection; we do not retain raw request payloads beyond the JSON parameters needed to debug a misbehaving query.

What's next

The public endpoint went live in May 2026. We are watching usage data and the MCP ecosystem closely. Likely next additions: a tool for cohort statistics on Canadian medical schools (the heterogeneous data ingested under our Canadian medical pipeline), a tool for the inverse calculator ("given a target acceptance category, what test score do I need"), and OAuth-based authentication so the personalised tools can be unlocked without a manually pasted Bearer token. If you build something interesting against AdmitBase MCP, send it to support@admitbase.com — we will feature good integrations in this article.

Discovery

AdmitBase publishes a discovery manifest at https://admitbase.com/.well-known/mcp.json for AI clients and registries that auto-detect MCP servers. The shorter install hub at https://admitbase.com/mcp has copy-pasteable config for the common AI assistants. This article is the canonical write-up; bookmark or cite it.