DeveshB-1/agentfile
Dockerfile for AI agents. Define your model, tools, memory, and system prompt in a single file.
Platform-specific configuration:
{
"mcpServers": {
"agentfile": {
"command": "npx",
"args": [
"-y",
"agentfile"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
<div align="center">
Dockerfile for AI agents.
Define your agent's model, tools, memory, and system prompt in a single file. Run it anywhere.
FROM claude-sonnet
SYSTEM "You are an expert software engineer."
TOOL ghcr.io/modelcontextprotocol/server-filesystem:latest filesystem
TOOL ghcr.io/modelcontextprotocol/server-github:latest github
MEMORY sqlite://./memory.dbagentfile runThat's it.
<br />
[](https://opensource.org/licenses/MIT) [](https://github.com/DeveshB-1/agentfile/stargazers)
</div>
---
Every AI agent project today is a mess of scattered config, hardcoded prompts, and manual setup steps. There's no standard way to define what an agent is.
Docker solved this for containers with a single Dockerfile. Agentfile does the same for AI agents.
---
curl -fsSL https://raw.githubusercontent.com/DeveshB-1/agentfile/main/install.sh | shOr with Go:
go install github.com/DeveshB-1/agentfile/cmd/agentfile@latest---
# Base model
FROM claude-sonnet
# System prompt (single line)
SYSTEM "You are a helpful assistant."
# System prompt (multi-line)
SYSTEM """
You are an expert software engineer.
You write clean, idiomatic code.
You ask clarifying questions when requirements are unclear.
"""
# MCP tools — any container image
TOOL ghcr.io/modelcontextprotocol/server-filesystem:latest filesystem
TOOL ghcr.io/modelcontextprotocol/server-github:latest github
TOOL ghcr.io/modelcontextprotocol/server-brave-search:latest braveLoading reviews...