# sapiom/sapiom-js [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/sapiom/sapiom-js  
**GitHub Stars:** 21  
**npm Downloads (last month):** 1884  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/mcp-322

## Description
Sapiom MCP server — authentication, verification, and API tools

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "sapiom-js": {
    "command": "npx",
    "args": ["-y","@sapiom/agent-studio@latest"]
  }
}
```

## Documentation

## What sapiom/sapiom-js does

The sapiom/sapiom-js MCP server provides a local developer interface for building and operating agents on Sapiom. It is implemented by the `@sapiom/mcp` package and is intended for use from a coding agent. The server complements the Sapiom TypeScript SDK, which also includes packages for defining typed agent graphs, calling Sapiom capabilities, and running command-line workflows.

An agent in the Sapiom model is a typed graph of steps. Each step performs work and returns a directive that determines the next step. The surrounding SDK includes helpers such as `defineAgent`, `defineStep`, `goto`, and `terminate`, while the MCP layer gives a coding agent a way to work with the resulting projects.

## How it works

The sapiom/sapiom-js MCP server runs locally over the standard MCP command pattern.

```sh
claude mcp add sapiom-dev -- npx -y @sapiom/mcp
```

After registration, a coding agent can use the developer MCP to scaffold an agent, check its project, deploy it, and inspect it. The local surface is named `sapiom-dev`, and its operations use the `sapiom_dev_*` namespace.

This server should not be confused with Sapiom’s remote capability MCP. The local server is for development and agent operations. Remote capability services are used for paid tool calls such as sandboxes, repositories, coding models, search, and file storage. The repository documentation identifies these as two distinct MCP surfaces.

## Setup and configuration

Install and register the MCP server with an MCP-compatible coding client using the package runner command above, or configure the equivalent command in another client. The repository also provides a CLI path for the same development lifecycle:

```sh
npx @sapiom/cli agents init my-app
cd my-app
npx @sapiom/cli agents check
npx @sapiom/cli agents deploy
```

The SDK is written in TypeScript and organized as a monorepo. The MCP package is `@sapiom/mcp`; it is separate from `@sapiom/tools`, the typed client used by application code to call Sapiom capabilities.

## Tools and capabilities

The sapiom/sapiom-js MCP server supports the developer workflow described in the repository:

- Scaffold a Sapiom agent project.
- Validate an agent locally, including its bundle, manifest, and graph.
- Deploy an agent to Sapiom.
- Inspect or operate Sapiom agents from a coding agent.

For programmatic capability access outside MCP, `@sapiom/tools` creates an authenticated client with `SAPIOM_API_KEY`. Its documented examples create repositories, run coding models against a repository, use sandboxes, and push changes. Those capabilities belong to the SDK client and remote Sapiom services, not necessarily to the local developer MCP.

## Limitations and notes

The repository labels the SDK as beta and places it in the v0.x range, so APIs may change before version 1.0. The local MCP is aimed at development workflows rather than being described as the remote paid-capability service. Use the local server to help a coding agent build and operate Sapiom agents; select the remote capability MCP when the requirement is to expose paid Sapiom tools.

The README names Claude Code as a supported setup target. It does not provide a complete per-tool schema or enumerate every `sapiom_dev_*` operation, so tool-level behavior should be verified against the package documentation and the installed version.

_Full upstream README: https://allmcps.com/mcp/mcp-322/readme_