Skip to main content

MCP Servers

MCP (Model Context Protocol) servers add capabilities to your agent — GitHub integration, file operations, custom tool services. The agent discovers and uses these tools automatically once they're configured.

Configure in config.yaml​

YAML is the recommended way to declare MCP servers, since the config is structured:

mcp_servers:
  - name: github
    url: https://api.githubcopilot.com/mcp/
    headers:
      Authorization: "Bearer ${GITHUB_PAT}"

  - name: local-tools
    url: http://localhost:8080/mcp

Fields:

  • name (optional) — human-readable identifier
  • url (required) — MCP server endpoint URL
  • headers (optional) — request headers, useful for auth tokens

YAML supports ${VAR} substitution for secrets, so your token stays in .env rather than being committed.

Example: GitHub MCP Server​

The official GitHub MCP server gives the agent access to repositories, issues, and pull requests.

1. Generate a Personal Access Token​

In GitHub: Settings → Developer settings → Personal access tokens → Tokens (classic) → Generate new token.

Recommended scopes:

  • repo — full control of private repositories
  • read:org — read org and team membership
  • read:user — read user profile data

2. Add the Token to .env​

# .env
GITHUB_PAT=ghp_xxxxxxxxxxxxxxxxxxxx

3. Add the Server to config.yaml​

mcp_servers:
  - name: github
    url: https://api.githubcopilot.com/mcp/
    headers:
      Authorization: "Bearer ${GITHUB_PAT}"

4. Use It​

just demo "List the recent issues in my repository"
just demo "Show me pull requests that need review"

Reconnection Behavior​

The base agent wraps the underlying Strands agent in a ReconnectingAgentProxy that recreates the agent if an MCP connection drops mid-request. Failed requests retry once before surfacing a mcp_reconnect_failed error to the caller. This behavior is in server.py and applies to both sync and streaming endpoints.