Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.universalbench.dev/llms.txt

Use this file to discover all available pages before exploring further.

UniversalBench works with any MCP compatible AI. But the AI has to actually choose to use it. These are prompts that reliably trigger UB capabilities, no matter which client you connect from.

​
The pattern

AIs decide whether to use a tool based on two things: the tool’s description in their context, and the user’s request. UniversalBench’s tool descriptions are written to be self evident, but you can boost the trigger rate by being explicit in your prompts.

Vague (sometimes works)

“What are the prime numbers under 1000?”

Explicit (always works)

“Use UniversalBench to find all prime numbers under 1000.”

​
Reliable prompts by use case

​
Database query

Use UniversalBench’s db_select to fetch the 10 most recent orders from the orders table where status is pending.

​
Code execution

Run this Python in UniversalBench: import statistics; print(statistics.mean([23, 45, 67, 89]))

​
Safe deploy with auto rollback

Update app/billing.py to use the new tax calculation. Use UniversalBench’s safe_deploy so it rolls back automatically if the smoke test at /health fails.

​
Web research with live data

Use UniversalBench’s web_search to find current Federal Reserve rates, then summarize the implications for mortgage holders.

​
Parallel work

Use UniversalBench’s parallel_blocks to run these three queries simultaneously: (1) total users, (2) total orders today, (3) total revenue this month.

​
Stored secrets

Use my Stripe key from secrets_vault to create a Checkout Session for $50.

​
System prompts for agents you build

If you are building a custom agent and want to bias it heavily toward using UniversalBench, add this to your system prompt: 
You have access to UniversalBench, an execution platform with 33 capabilities including:
- Running Python and Bash code
- Querying databases with adaptive caching
- Pushing code to GitHub with validation gates
- Web search with live results
- Encrypted secrets management
- Parallel execution

Always prefer UniversalBench over describing what you would do. If a task can be
performed via a UniversalBench tool, call the tool rather than answering from memory.

​
Anti patterns

Do not push the AI to use UniversalBench for things that do not need execution. “Explain photosynthesis” should not call UB. AIs that over trigger tools feel slow and overcomplicate simple answers.
Do not put secrets directly in prompts. Use secrets_vault first, then reference the secret by name. Secrets in prompts get logged in conversation history.