Skip to content

MCP Server Integration

@agentic/triage includes a Model Context Protocol (MCP) server, enabling integration with Claude Desktop, Cursor, and other MCP-compatible clients.

The Model Context Protocol is an open standard for connecting AI models to external tools and data sources. With MCP, you can give Claude Desktop or Cursor access to triage tools without writing any code.

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
"mcpServers": {
"triage": {
"command": "npx",
"args": ["@jbcom/agentic", "mcp-server"]
}
}
}

Add to your Cursor MCP configuration:

{
"mcpServers": {
"triage": {
"command": "npx",
"args": ["@jbcom/agentic", "mcp-server"]
}
}
}
{
"mcpServers": {
"triage": {
"command": "npx",
"args": ["@jbcom/agentic", "mcp-server"],
"env": {
"GH_TOKEN": "your-github-token",
"GITHUB_REPOSITORY": "owner/repo"
}
}
}
}

Once configured, these tools are available to the AI:

ToolDescription
list_issuesList issues with filtering
get_issueGet a specific issue
create_issueCreate a new issue
update_issueUpdate an existing issue
close_issueClose an issue
search_issuesSearch issues by query
add_labelsAdd labels to an issue
remove_labelsRemove labels from an issue
ToolDescription
get_pr_commentsGet comments on a PR
add_pr_commentAdd a comment to a PR
approve_prApprove a pull request
request_changesRequest changes on a PR
ToolDescription
get_sprintsGet all sprints
get_current_sprintGet the current sprint
get_sprint_issuesGet issues in a sprint
move_to_sprintMove issue to a sprint

After configuring the MCP server, you can ask Claude:

“List all open bugs in my repository and prioritize them by severity”

Claude will use the list_issues and search_issues tools to fetch issues and provide analysis.

“Create an issue for the login bug we discussed, label it as ‘critical’ and assign it to @developer”

Claude will use create_issue and add_labels to create the issue.

“Review PR #123 and add a comment about the authentication changes”

Claude will use get_pr_comments and add_pr_comment to review and comment.

In Cursor’s AI chat, you can:

“Triage all open issues and create a sprint plan”

Cursor will use the triage tools to analyze issues and suggest a sprint plan.

If you prefer to install globally:

Terminal window
npm install -g @jbcom/agentic

Then configure:

{
"mcpServers": {
"triage": {
"command": "agentic-triage",
"args": ["mcp-server"]
}
}
}
{
"mcpServers": {
"triage": {
"command": "npx",
"args": ["@jbcom/agentic", "mcp-server", "--provider", "github"],
"env": {
"GH_TOKEN": "your-token",
"GITHUB_REPOSITORY": "owner/repo"
}
}
}
}
{
"mcpServers": {
"triage-jira": {
"command": "npx",
"args": ["@jbcom/agentic", "mcp-server", "--provider", "jira"],
"env": {
"JIRA_HOST": "your-company.atlassian.net",
"JIRA_EMAIL": "your-email@company.com",
"JIRA_API_TOKEN": "your-api-token",
"JIRA_PROJECT": "PROJ"
}
}
}
}
{
"mcpServers": {
"triage-linear": {
"command": "npx",
"args": ["@jbcom/agentic", "mcp-server", "--provider", "linear"],
"env": {
"LINEAR_API_KEY": "your-api-key",
"LINEAR_TEAM_ID": "your-team-id"
}
}
}
}

For development or testing:

Terminal window
# Run the MCP server
npx @jbcom/agentic mcp-server
# With debug logging
DEBUG=triage:* npx @jbcom/agentic mcp-server
# With specific provider
npx @jbcom/agentic mcp-server --provider github
Terminal window
# Test the server starts correctly
npx @jbcom/agentic mcp-server --test
# List available tools
npx @jbcom/agentic mcp-server --list-tools

In Claude Desktop, check the logs at:

  • macOS: ~/Library/Logs/Claude/
  • Windows: %APPDATA%\Claude\logs\

Server not starting:

Terminal window
# Check Node.js version (requires 18+)
node --version
# Reinstall the package
npm install -g @jbcom/agentic

Authentication errors:

Terminal window
# Verify your token works
GH_TOKEN=your-token gh api user

Repository not found:

Terminal window
# Ensure GITHUB_REPOSITORY is set correctly
export GITHUB_REPOSITORY="owner/repo"

For GitHub, your token needs:

  • repo - Full repository access
  • read:org - Read organization data (optional)
  1. Use fine-grained tokens - Create tokens with minimal necessary permissions
  2. Rotate regularly - Update tokens periodically
  3. Don’t commit tokens - Use environment variables or secure storage
  4. Review tool calls - Claude will ask before executing destructive operations