Skip to content
ERA Agent

MCP Quick Reference

Quick reference for using ERA Agent with Claude Desktop via MCP.

Setup (One-Time)

Section titled “Setup (One-Time)”
Terminal window
# 1. Build agent
cd era-agent && go build -o agent


# 2. Configure Claude Desktop
# Edit: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
# Edit: ~/.config/Claude/claude_desktop_config.json (Linux)
{
  "mcpServers": {
    "era-agent": {
      "command": "/absolute/path/to/agent",
      "args": ["mcp"]
    }
  }
}
Terminal window
# 3. Restart Claude Desktop (Cmd+Q, then reopen)

Quick Tool Reference

Section titled “Quick Tool Reference”

Language Tools (No Session Required)

Section titled “Language Tools (No Session Required)”
ToolUse ForExample
era_pythonPython scriptsRun FizzBuzz, data analysis
era_nodeJavaScript/Node.jsArray operations, algorithms
era_typescriptTypeScript codeType-safe code execution
era_denoDeno runtimeModern JS/TS with Deno APIs
era_shellSystem commandsPackage install, file ops

Session Tools (For Stateful Work)

Section titled “Session Tools (For Stateful Work)”
ToolUse ForExample
era_create_sessionStart persistent environmentCreate Python session
era_run_in_sessionRun code in sessionExecute with state
era_list_sessionsView all sessionsCheck active VMs
era_get_sessionGet session detailsCheck session status
era_delete_sessionClean up sessionRemove unused VM

File Operations (Session Only)

Section titled “File Operations (Session Only)”
ToolUse ForExample
era_upload_fileAdd files to sessionUpload CSV, config
era_read_fileRead session filesView results
era_list_filesList session filesCheck workspace

Common Patterns

Section titled “Common Patterns”

Quick Python Script

Section titled “Quick Python Script”
Run this Python code:
print("Hello World")
print([x**2 for x in range(10)])

Install Packages

Section titled “Install Packages”
1. Create a Python session
2. In that session, run: pip install pandas requests
3. Then run your Python code using those packages

Multi-Step Analysis

Section titled “Multi-Step Analysis”
1. Create a Python session
2. Upload data.csv with [your data]
3. Install pandas
4. Analyze the data
5. Read the results file

Multi-Language Testing

Section titled “Multi-Language Testing”
Test this sorting algorithm in both Python and JavaScript

Web Scraping

Section titled “Web Scraping”
Create a Python session with network access, install beautifulsoup4 and requests,
then scrape the top stories from Hacker News

Gotchas

Section titled “Gotchas”

✅ Works Great

Section titled “✅ Works Great”
  • F-strings: print(f'{x}: {y}')
  • Multi-line code with proper indentation
  • Special characters in strings
  • Complex nested structures

⚠️ Watch Out

Section titled “⚠️ Watch Out”
  • No State Between Tools: Use sessions for stateful work
  • Packages Not Installed: Use era_shell to install first
  • Network Disabled: Request “with network access” when needed
  • File Paths: Use relative paths in sessions, upload files first

Default Limits

Section titled “Default Limits”
ResourceDefaultOverride
CPU1 core”with 4 CPUs”
Memory256 MiB”with 2GB memory”
Timeout30 sec”with 60 second timeout”
NetworkDisabled”with network access”

Troubleshooting

Section titled “Troubleshooting”
ProblemSolution
Tools not showingRestart Claude Desktop (Cmd+Q)
“Command not found”Use absolute path in config
Code execution failsCheck Docker is running
Package not foundUse era_shell to install
Network errorRequest session “with network access”

Best Practices

Section titled “Best Practices”

When to Use What

Section titled “When to Use What”

Use Language Tools (era_python, etc.):

  • Quick one-off scripts
  • Simple calculations
  • Testing snippets
  • No package dependencies

Use Sessions:

  • Installing packages
  • Multi-step workflows
  • Stateful computations
  • File processing pipelines

Use era_shell:

  • Package installation (pip, npm)
  • System commands (ls, mkdir)
  • Environment setup
  • NOT for running code (use language tools instead)

Performance Tips

Section titled “Performance Tips”
  1. Reuse Sessions: Create once, use multiple times
  2. Choose Right Tool: Language-specific tools are faster than generic
  3. Install Once: Install packages at session start, not each run
  4. Load Once: Load large data once in session, process multiple times

Example Prompts

Section titled “Example Prompts”

Data Science

Section titled “Data Science”
Create a Python session, install pandas and matplotlib, upload this CSV,
analyze the data, create a visualization, and save it as a PNG

Algorithm Development

Section titled “Algorithm Development”
Implement quicksort in Python with logging, test it with random data,
then implement the same algorithm in JavaScript and compare performance

API Integration

Section titled “API Integration”
Create a Python session with network access, install requests,
fetch data from this API, process it, and save as JSON

Multi-File Project

Section titled “Multi-File Project”
Create a Node.js session, create package.json, install dependencies,
create index.js with an Express server, and test it

Security Notes

Section titled “Security Notes”
  • ✅ All code runs in isolated VMs
  • ✅ Network disabled by default
  • ✅ Filesystem isolation per session
  • ✅ Resource limits enforced
  • ⚠️ Enable network only when needed
  • ⚠️ Sessions are local to your machine
Section titled “Quick Links”