# Agent University

> Connect to and use the Agent University hosted MCP server — evidence-backed knowledge (build guides, gotchas, recipes, code samples) for building with LLM-tooling targets. Use when building with a tool, SDK, or API covered by an Agent University degree, when you need a step-by-step build guide grounded in live-verified evidence, or when debugging a platform-specific error.

- Skill: `dennisonbertram/agent-university` (Agent Skill)
- Install (CLI): `npx skillmds@latest add dennisonbertram/agent-university`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dennisonbertram/agent-university/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: dennisonbertram (https://skillmd.com/u/dennisonbertram)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dennisonbertram/agent-university

---


# Agent University MCP Server

Agent University is a knowledge corpus built from real, live-verified builds
("degrees") of LLM-tooling targets — SDKs, APIs, CLIs, and cloud services.
Every artifact carries an evidence tier and a source citation back to the build
that produced it. The corpus is served by a hosted MCP server.

## Requirements

- An Agent University API key in the `AGENT_UNIVERSITY_API_KEY` environment
  variable. The server rejects unauthenticated requests. Keys are issued by the
  maintainer — request one via an issue on this repo.

## Register the server

Claude Code:

```bash
claude mcp add --transport http agent-university \
  https://agent-university-api-production.up.railway.app/v1/mcp \
  --header "Authorization: Bearer ${AGENT_UNIVERSITY_API_KEY}"
```

Any other MCP client (streamable HTTP):

```json
{
  "url": "https://agent-university-api-production.up.railway.app/v1/mcp",
  "headers": { "Authorization": "Bearer <AGENT_UNIVERSITY_API_KEY>" }
}
```

## How to use the tools

**Building with a covered target? Call `agent_university_how_to_build` FIRST.**
It returns the complete, ordered build guide assembled from how the target was
actually built and verified — verbatim source content with citations, not
search snippets.

Knowledge tools:

| Tool | When to use |
|---|---|
| `agent_university_how_to_build` | First call for any "how do I build X" task — full ordered guide. |
| `agent_university_search` | General queries; supports target/degree/artifact/evidence filters. |
| `agent_university_plan_build` | Planning context: recipes, gotchas, instructions for a target. |
| `agent_university_get_gotchas` | Before writing code against a target — known traps, live-verified. |
| `agent_university_get_recipe` | Working recipes for a target or degree. |
| `agent_university_debug_error` | Hit a platform-specific error or symptom — search known fixes. |
| `agent_university_get_artifact` | Fetch a full artifact by ID (from search results). |
| `agent_university_get_code_sample` | Fetch a code sample by ID with source citation. |
| `agent_university_get_degree_status` | Degree metadata for a target. |

Ops tools (`agent_university_ops_*`: deployment, checks, freshness, probe,
metrics, events, runbooks, runbook, error_catalog) report on the hosted API
itself — corpus version, health checks, retrieval mode, error-code handling.
Use them to verify the server is current before trusting results, or when a
call fails: `ops_freshness` answers "is the corpus live and current?" in one
call; `ops_error_catalog` says whether an error is retryable.

Recommended flow: `how_to_build` (or `search`) → `get_gotchas` before coding →
`debug_error` when something breaks. Prefer artifacts marked with higher
evidence tiers (e.g. `live_verified`) when results conflict.