Anter MCP Server
Anter exposes its own Agent Builder platform as an MCP server. Connect an MCP-compatible client — Claude Code, Claude Desktop, OpenCode, Cursor, or any agent CLI that speaks the protocol — and it gains direct, structured access to your organization's agents, projects, model providers, and connected MCP servers.
Instead of clicking through the dashboard, you (or an AI coding assistant working on your behalf) can design an agent, iterate on its spec, run it, and inspect the result — all from the terminal.
Not the same as connecting an MCP server to your agents
This section covers Anter acting as an MCP server for your CLI or IDE. If you're looking to connect a third-party MCP server (Tavily, a vector database, an internal API) so that your agents can call its tools, see MCP servers under Flows instead. Both features are unrelated to each other beyond sharing the same protocol.
Quickstart
Get your API key and connect your first MCP client in a few minutes.
Key concepts
Authentication, transports, sessions, and how scoping keeps your data isolated.
Agent lifecycle tools
Design, revise, and inspect agents through MCP tool calls.
What you can do with it
| Capability | Example |
|---|---|
| Design and iterate on agents | Ask your CLI agent to draft an AgentSpec, create it as a draft, then revise it based on feedback — without leaving the terminal |
| Inspect a live agent | Look up an agent by slug, read back its full graph, or fetch the JSON Schema the spec format follows |
| Run and observe | Trigger a real run of a draft or live agent, then pull the resulting session history or routing trace |
| Manage projects and providers | List projects, add or update LLM providers, and check what fields a given provider type requires |
| Manage connected MCP servers | List, add, update, or test the MCP servers already connected to your projects |
| Query skills and toolsets | Browse the skill catalog and Anter's builtin toolset catalog before wiring them into a spec |
Why use it instead of the dashboard
The Anter dashboard is built for point-and-click configuration. The MCP server is built for conversational, scripted, and agentic workflows — an AI coding assistant that already has your codebase open can design an agent spec informed by your actual API contracts, or iterate on a spec across several turns without you re-explaining context each time.
The two surfaces share the same backend: an agent created via create_draft_agent shows up in the dashboard immediately, and vice versa.
New to Anter's agent model?
If you haven't built an agent before, skim Key concepts in Flows first — the MCP tools operate on the same agents, projects, and specs described there.