Nestor MCP Server

The @nestor/mcp package exposes Nestor's agent, workflow, and skill surface as a Model Context Protocol server. Any MCP-compatible client can list agents, kick off workflows, or install skills without touching the CLI.

Overview

The Model Context Protocol (MCP) is Anthropic's open standard for connecting LLM clients to external tools and data. Nestor ships a first-party MCP server that wraps the same primitives used by the CLI and Studio — agents, workflows, skills — so you can drive Nestor directly from Claude Desktop, Cursor, Continue, Cline, or any client that speaks MCP over stdio.

The server is implemented in packages/mcp/src/server.ts with tool modules under packages/mcp/src/tools/. It shares the NestorStore SQLite database with the rest of the stack, so a workflow kicked off from Claude Desktop shows up in Studio the moment you refresh.

Why MCP matters

Without MCP, using Nestor from an IDE means shelling out to npx nestor-sh and copy-pasting output back. MCP turns Nestor into a set of first-class tools inside the assistant:

The 15 tools

The server currently exposes 15 tools across three functional groups. The canonical list is assembled in packages/mcp/src/tools/index.ts by concatenating agentTools, workflowTools, skillTools, and metaTools.

Agent ops (5)

ToolPurpose
nestor_list_agentsList all registered agents. Filter by trust level.
nestor_create_agentCreate a new agent with an LLM adapter (claude, openai, ollama…) and model.
nestor_run_agentRun an agent with a prompt and return the response.
nestor_get_agent_statusInspect an agent's current run state and recent activity.
nestor_delete_agentRemove an agent from the store.

Workflow ops (4)

ToolPurpose
nestor_list_workflowsList registered workflow definitions.
nestor_create_workflowRegister a new workflow (name + description).
nestor_run_workflowStart an execution with optional JSON-encoded input.
nestor_get_run_statusPoll the state of a running or completed workflow.

Skill ops (6)

ToolPurpose
nestor_list_skillsList installed skills, optionally filtered by tag.
nestor_install_skillInstall a skill from a local path or a git URL (depth 1, hooks disabled).
nestor_scan_skillRun the static-analysis scanner on a skill before trusting it.
nestor_build_agentMeta-tool: scaffold a new agent from a prompt description.
nestor_build_skillMeta-tool: scaffold a new skill (SKILL.md + tests) from a spec.
nestor_translate_skillConvert a skill between Nestor, MCP, LangChain, CrewAI, and OpenAI function formats.

Install in Claude Desktop

Claude Desktop loads MCP servers from claude_desktop_config.json. On macOS the file lives at ~/Library/Application Support/Claude/claude_desktop_config.json; on Windows it lives at %APPDATA%\Claude\claude_desktop_config.json.

Add Nestor to the mcpServers object (the npm package is nestor-sh, not nestor):

{
  "mcpServers": {
    "nestor": {
      "command": "npx",
      "args": ["-y", "nestor-sh", "mcp"],
      "env": {
        "NESTOR_HOME": "~/.nestor"
      }
    }
  }
}

Restart Claude Desktop. The 15 tools appear under the plug icon. Ask Claude to "list Nestor agents" to confirm the handshake.

Heads up: the MCP server writes to the same SQLite store as the CLI. Don't run npx nestor-sh start and the MCP server against the same NESTOR_HOME on Windows without WAL mode enabled — concurrent writers will serialize.

Other clients

For more on the underlying primitives, see the documentation overview, or dive into agents, workflows, and skills.