Decionis MCP server
{
"mcpServers": {
"decionis": {
"command": "npx",
"args": [
"-y",
"@decionis/mcp@latest"
]
}
}
}{
"mcpServers": {
"decionis": {
"command": "npx",
"args": [
"-y",
"@decionis/mcp@latest"
]
}
}
}{
"mcpServers": {
"decionis": {
"command": "npx",
"args": [
"-y",
"@decionis/mcp@latest"
]
}
}
}{
"servers": {
"decionis": {
"command": "npx",
"args": [
"-y",
"@decionis/mcp@latest"
]
}
}
}{
"mcpServers": {
"decionis": {
"command": "npx",
"args": [
"-y",
"@decionis/mcp@latest"
]
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y @decionis/mcpversion 0.3.0
What it does
Decionis is an MCP server that implements an authorization boundary for AI agent actions. It lets agents propose actions while a separate policy engine decides whether to allow, escalate to a human, or block each action before execution.
- Require human approval for high-risk agent actions like refunds or deployments before they execute.
- Enforce budget or capability limits on agent-proposed actions based on policy rules.
- Prevent agents from directly accessing downstream credentials or choosing which handlers execute their actions.
Tools this server exposes
| Tool | Does |
|---|---|
| IntentCapture | Captures and canonically hashes an agent's proposed action with trusted context. |
| DecionisGate | Evaluates a captured intent against policy and returns an ALLOW, ESCALATE, or BLOCK decision. |
| SafeExecutor | Executes an action only if it has a valid decision and a single-use grant bound to that intent. |
| ActionRegistry | Maps action names to trusted handlers, validates parameters, and consumes grants before execution. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.