Safe-Bifrost
A local MCP server that provides a safe plan-and-execute workflow for AI coding assistants, storing plans and tasks, and executing agent commands with an allow-list for security.
README
Safe-Bifrost
Safe-Bifrost is a local Model Context Protocol (MCP) server for a safe plan-and-execute workflow:
- An MCP client saves an implementation plan.
- Safe-Bifrost stores the plan inside one configured workspace.
- A local runner executes the task with an allow-listed local agent command.
- The client reads back the result, git diff, test log, and task status.
It is designed for local AI coding workflows where ChatGPT, Claude, Codex, OpenCode, or another MCP client should not receive arbitrary shell access.
Features
- MCP stdio server with 9 tools.
- Workspace-scoped plan and task storage under
.safe-bifrost/. - Path traversal and symlink containment checks.
- Sensitive file read blocking for
.env, SSH keys, tokens, credentials, browser cookies, npm credentials, Kubernetes config, and similar files. - Agent command allow-list through
safe-bifrost.config.json. - Test command allow-list with exact command matching.
- Local runner that captures
result.md,git.diff, andtest.log. - Windows-friendly Node.js scripts.
Tools
Safe-Bifrost exposes these MCP tools:
save_planget_plancreate_taskget_task_statusget_resultget_diffget_test_loglist_workspaceread_workspace_file
Requirements
- Node.js 18 or newer.
- npm for local development.
- A local agent command if you want runner execution, such as
codexoropencode.
Install From Source
Windows PowerShell:
cd D:\ai_agent\Reasonix\reasonix_program\safe-bifrost
npm.cmd install
npm.cmd run build
npm.cmd run test:mcp
Linux, WSL, or Git Bash:
cd safe-bifrost
npm install
npm run build
npm run test:mcp
Configure
Create safe-bifrost.config.json. Save it as UTF-8. The server also accepts
UTF-8 with BOM, but UTF-8 without BOM is recommended for portability.
{
"workspaceRoot": "D:/ai_agent/my-project",
"plansDir": ".safe-bifrost/plans",
"tasksDir": ".safe-bifrost/tasks",
"agents": {
"codex": {
"command": "codex",
"args": ["exec", "--cd", "{repo}", "{prompt}"]
},
"opencode": {
"command": "opencode",
"args": ["run", "{prompt}"]
}
},
"allowedTestCommands": [
"npm test",
"npm run test",
"pytest",
"cargo test"
],
"maxReadFileBytes": 200000
}
Important fields:
workspaceRoot: absolute path to the workspace that Safe-Bifrost may read and write.plansDir: plan storage directory, relative toworkspaceRoot.tasksDir: task storage directory, relative toworkspaceRoot.agents: allow-listed local agent commands. The{repo}and{prompt}placeholders are replaced by the runner and passed as process arguments.allowedTestCommands: exact test commands that clients may request.maxReadFileBytes: maximum bytes returned by file-reading tools.
Run The MCP Server
Windows PowerShell:
$env:SAFE_BIFROST_CONFIG = "D:\ai_agent\Reasonix\reasonix_program\safe-bifrost\safe-bifrost.config.json"
node dist\index.js
Linux, WSL, or Git Bash:
SAFE_BIFROST_CONFIG=/path/to/safe-bifrost.config.json node dist/index.js
MCP Client Configuration
Use an absolute path for both the server entrypoint and config file.
{
"mcpServers": {
"safe-bifrost": {
"command": "node",
"args": [
"D:/ai_agent/Reasonix/reasonix_program/safe-bifrost/dist/index.js"
],
"env": {
"SAFE_BIFROST_CONFIG": "D:/ai_agent/Reasonix/reasonix_program/safe-bifrost/safe-bifrost.config.json"
}
}
}
}
Workflow
-
Call
save_planwith a title and Markdown plan. -
Call
create_taskwith the returnedplan_id, an allow-listedagent, and optionally an allow-listedtest_command. -
Run the task locally:
npm.cmd run runner -- task_xxx -
Call
get_result,get_diff, andget_test_logto review outputs.
Security Model
Safe-Bifrost intentionally avoids general shell execution through MCP tools.
- Clients cannot pass arbitrary shell commands.
- Agent commands must be configured ahead of time.
- Test commands must match
allowedTestCommandsexactly. - File reads are contained to
workspaceRoot. - Sensitive files are blocked even when they are inside the workspace.
- The runner does not commit, push, delete files, or reset repositories by itself.
This project is still a local automation bridge, so configure workspaceRoot
and agents carefully.
Development Commands
Windows PowerShell:
npm.cmd run build
npm.cmd test
npm.cmd run test:mcp
npm.cmd run verify:package
npm.cmd run pack:clean
test:mcp starts the MCP server over stdio, calls the real tools, verifies
security rejections, and runs the local runner with a harmless placeholder
agent command.
Release Artifacts
Generate a clean source/dist archive:
npm.cmd run pack:clean
Generate the npm package tarball:
npm.cmd pack
The clean release archive excludes node_modules/, .safe-bifrost/, logs,
local config files, and .env.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.