AI agent support

StackNext is designed for agents that need runtime evidence before changing code, and for agents that must use the APIs of the installed version rather than the latest documentation on the internet.

The four pieces

Piece	What it gives the agent	Where it lives
Package docs	Version-matched API reference in installed packages	`node_modules/@stacknext/*/docs/`
Managed `AGENTS.md`	A pointer to local docs and evidence rules	Project `AGENTS.md`, installed by `stacknext init --agents`
Agent Skill	Debugging, diagnostics, regression and snapshot workflows	Client skill directory, installed by `stacknext skill install`
MCP server	Read-only access to live telemetry and saved records	`stacknext mcp` over stdio

The pieces are intentionally separate. Package docs contain technical reference; the skill contains procedures; the MCP server contains runtime data; AGENTS.md carries the short project-level rules.

The evidence contract

An agent using StackNext must:

Start with get_project_summary and confirm the app is connected.
Cite routes, trace ids, error ids, spans, durations, statuses or finding ids.
Treat diagnostic findings as evidence, not confirmed root causes.
Treat inconclusive comparisons as inconclusive, not passes.
Say when a result comes from a snapshot instead of the live sample.
Respect the redaction boundary and never try to recover removed values.
Never claim support for an API, config key or package that is not in the installed docs.

Start here

Setup — install instructions, the skill and MCP clients.
Evidence-first workflows — the shared investigation loop.
Local documentation — version-matched files and AI-readable endpoints.
MCP server — start the server and review its security model.

Overview

Getting started

Guides

Integrations

AI agents

Reference

Troubleshooting

Security and privacy

The four pieces

The evidence contract

Start here

​The four pieces

​The evidence contract

​Start here

The four pieces

The evidence contract

Start here