MCPs (Model Context Protocol)

Extend Zenflow agents with additional tools and integrations using MCPs.

Overview

Model Context Protocol (MCP) is an open standard that allows AI agents to interact with external tools and services. Zenflow supports MCPs to extend agent capabilities beyond code editing.

What MCPs enable

Give agents access to databases, APIs, file systems, development tools, and custom integrations—all through a standardized protocol.

Built-in MCPs

Zenflow includes several MCPs out of the box:

MCP	Capabilities
Git	Clone, commit, push, branch, merge
File System	Read, write, search files
Terminal	Execute shell commands
Browser	Navigate, interact with web pages
Search	Search code across repositories

Adding Custom MCPs

From the MCP Library

1. Go to Settings → Agent Tools
2. Browse the MCP library
3. Click Install on the desired MCP
4. Configure any required settings (API keys, endpoints)

Manual Configuration

For custom or private MCPs:

1. Go to Settings → Agent Tools → Add Custom
2. Enter the MCP configuration:

{
  "name": "my-custom-mcp",
  "command": "npx",
  "args": ["-y", "@myorg/mcp-server"],
  "env": {
    "API_KEY": "${MY_API_KEY}"
  }
}

3. Save and restart the agent

Popular MCPs

Database

Query PostgreSQL, MySQL, SQLite directly from agents

Jira

Create issues, update status, link to tasks

Slack

Send notifications, read channel history

AWS

Manage S3, Lambda, EC2 resources

MCP Permissions

Control what agents can access:

Permission Level	Description
Read-only	Agent can query but not modify
Read-write	Full access to the tool
Approval required	Human must approve each action
Disabled	MCP available but not active

Configure in Settings → Agent Tools → [MCP Name] → Permissions.

Security Considerations

MCPs grant agents real-world capabilities. Review permissions carefully:

• Limit write access to necessary tools only
• Use approval mode for sensitive operations
• Rotate API keys regularly
• Audit agent actions in chat logs

Sandboxing

Zenflow runs MCPs in isolated environments:

• Separate process per MCP
• Limited system access
• Configurable resource limits
• Network restrictions available

Creating Custom MCPs

Build MCPs for your internal tools:

1. Implement the MCP server spec
2. Define available tools and their schemas
3. Handle tool invocations
4. Package and distribute

Resources:

• MCP Specification
• Example Servers

Troubleshooting

MCP not connecting

Verify that:

• The command path is correct
• Required dependencies are installed
• Environment variables are set
• Network access is available (if remote)

Agent not using the MCP

Check that:

• The MCP is enabled for the project
• The agent has permission to use it
• The tool is relevant to the current task

Actions failing

Review the error in the chat log. Common issues:

• Invalid API credentials
• Rate limiting
• Permission denied
• Malformed requests

Next Steps

GitHub Integration

Connect to GitHub

Orchestrating Agents

Configure agent behavior