Skip to main content
AgentMark plugs into AI coding agents and editors so they can author prompts, drive the full API, and debug traces without leaving your editor. It works with Claude Code, Cursor, Codex, GitHub Copilot, VS Code, Zed, and other Model Context Protocol and agentskills.io-compatible tools.

Start in one command

npm create agentmark@latest (or agentmark init inside an existing repo) wires every method below into the editors you select: it installs the agent skill and registers the docs and gateway MCP servers. Nothing else to configure.

Pick a method

There are three ways to connect an agent to AgentMark, plus the CLI. They all share the same gateway API, so pick the one that fits how your agent works. Most setups use several at once: the skill and docs MCP for authoring, the gateway MCP or CLI for driving the API.
MethodWhat it gives the agentReach for it when
Agent skillThe full AgentMark playbook (CLI commands, .prompt.mdx authoring, datasets, experiments, evals, git deploys), auto-loaded at session startYou want any agent to use AgentMark correctly, in any project
Docs MCPA live query interface over the AgentMark docs, so the agent authors files from current guidance instead of memoryThe agent is creating or editing prompts, datasets, or config
Gateway MCPOne MCP tool per gateway endpoint: traces, spans, datasets, scores, experiments, apps, deployments, environments, alertsThe agent speaks MCP and needs to drive or inspect AgentMark
CLIThe same gateway API from the shell, so the agent can pre-filter results before they reach its context windowThe agent can run bash
The skill and docs MCP help an agent author AgentMark correctly. The gateway MCP and CLI let it drive the API (query traces, manage deployments). They compose: a typical session uses the skill for conventions, the docs MCP for authoring, and the gateway MCP or CLI to inspect results.

Agent skill

Install the playbook so your agent already knows AgentMark

Docs MCP

Let your editor author files from the live docs

Gateway MCP

Drive the full AgentMark API from your editor

CLI commands

The same API from the shell, for bash-capable agents

Have questions?

Reach out any time: