Modelara helps engineering teams measure whether changes to agent-facing interfaces make AI agents better or worse at using them.
Modelara is currently focused on MCP interface assurance. It is being developed around a repeatable workflow for inspecting an interface, benchmarking agent behavior, comparing versions, and defining release criteria.
01 Get oriented
Quick Start
Start with the user guide for installation, project setup, and your first MCP regression workflow. The technical overview explains the workflow and CLI in more depth.
01
Inspect an agent-facing interface
Establish what the interface exposes to a model.
02
Benchmark representative tasks
Capture repeatable behavior as a baseline.
03
Compare a candidate change
Understand differences from the known baseline.
04
Define release criteria
Decide which behavioral changes should block a release.
02 Core workflow
From interface to evidence.
01 / INSPECT
Map the interface
Inspect the MCP interface exposed to an agent. modelara inspect lists the tools a server exposes, with their descriptions, schemas, declared and derived effects (read, write, destructive or unknown) and an interface fingerprint. modelara check --static compares two versions of an interface without a model or tool calls.
02 / BENCHMARK
Measure behavior
Run a repeatable suite of tasks against an interface and establish a baseline. Task success, tool calls, turns, token usage, latency, tool errors, and wrong answers are example signals under consideration—not a guarantee of current product output.
03 / COMPARE
Understand what changed
Compare a known baseline with a candidate interface version and review behavioral differences.
Baseline↓Candidate↓Behavioral differences
For interfaces, use modelara check --static; for agent behavior, modelara test compares against your baseline.
04 / CHECK
Set release criteria
Turn measured differences into a release decision: PASS, REVIEW, FAIL or INVALID, with stable exit codes for CI. See results and exit codes.