mcp · sapiom — Claude Code MCP server: build Sapiom agents
A local MCP server named sapiom-dev that lets a coding agent scaffold, validate and run Sapiom agents on your machine, then deploy and inspect them in the cloud.
claude mcp add mcp -- npx -y @sapiom/mcpWhat mcp does
The server runs over stdio on your own machine under the name sapiom-dev. sapiom_dev_agents_scaffold sets up a fresh agent project, sapiom_dev_agents_check validates types, imports, the bundle and the step graph, and sapiom_dev_agents_run_local executes the code locally. During a local run only ctx.sapiom.* calls get replaced with stubs; network access, filesystem writes and process calls in author code still happen for real.
Networked tools — link, deploy, run, inspect, signal — need a Sapiom API key. Trigger sapiom_authenticate once and a browser login opens; the key is then cached per environment in ~/.sapiom/credentials.json. Keys are sent over https:// or over plain http:// to a loopback host only, and a request to any other http:// target is refused before it leaves. Trusted private networks can opt out with SAPIOM_ALLOW_INSECURE_HTTP=1 in the server env.
This is not the hosted Sapiom MCP with capability tools. No capability tool lives here. Deploys, cloud builds, production runs, signals and schedules touch Sapiom cloud state and may be metered. The local tools scaffold, check and run_local need no authentication, though scaffold may query npm for current dependency versions.
What it helps with
- start a new Sapiom agent project from the agent
- catch type and step-graph errors before deploying
- run an agent locally without Sapiom capability spend
- publish an app to a live sandbox URL that expires with its
ttl - wait on a cloud execution or build and read the result
Example prompts for the agent
- Scaffold a new Sapiom agent project and list the generated files
- Check the project before deploy: types, bundle and step graph
- Run the agent locally and show what each step returned
How to install mcp
Platform CLI
claude mcp add mcp -- npx -y @sapiom/mcpPlatform CLI
codex mcp add mcp -- npx -y @sapiom/mcpConfig file
{
"mcpServers": {
"mcp": {
"command": "npx",
"args": [
"-y",
"@sapiom/mcp"
]
}
}
}Config file
{
"servers": {
"mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@sapiom/mcp"
]
}
}
}Platform CLI
gemini mcp add mcp npx -y @sapiom/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about mcp
How do I install mcp in Claude Code?
Run claude mcp add mcp -- npx -y @sapiom/mcp. There is no global install; the server starts on demand through npx.
What does mcp do?
It gives the coding agent tools to create, validate and locally run Sapiom agents, deploy them, inspect executions and manage triggers.
Which tools work without a Sapiom key?
scaffold, check and run_local do not need authentication. Networked calls such as link, deploy, run, inspect and signal need a key obtained through sapiom_authenticate.
Similar MCP servers
All MCP servers for Agent setup: memory, context, orchestration →mcp by other authors comes first.
