navra

Getting Started

Install

From source (Fedora/RHEL)

cargo install --path navra-server

From release binary

Download the latest binary from GitHub Releases and place it in your $PATH.

First-time setup

Run the interactive setup wizard:

navra init

The wizard walks you through:

  1. Agent detection — detects Claude Code, Goose, or custom agents
  2. Project type — dev, data, ops, or custom (drives MCP server recommendations)
  3. Model backend — Ollama, Mistral, Anthropic, OpenAI-compatible, or none
  4. Safety level — standard, strict, or minimal
  5. Directory access — which paths agents can read/write
  6. Systemd service — optional auto-start on login

It generates a config file, a bearer token, and prints connection instructions for your agent.

For scripted/CI use:

navra init --quiet --agent-name claude --project dev --safety standard

Connect a client

After navra init, connect your AI agent to the gateway.

Claude Code

Add to .claude/settings.json:

{
  "mcpServers": {
    "navra": {
      "url": "http://localhost:9315/mcp",
      "headers": {
        "Authorization": "Bearer mcd_your_token_here"
      }
    }
  }
}

Goose

Add to ~/.config/goose/config.yaml:

extensions:
  navra:
    type: sse
    uri: http://localhost:9315/mcp
    headers:
      Authorization: "Bearer mcd_your_token_here"

Any MCP client

navra speaks standard MCP over Streamable HTTP. Any MCP client can connect to http://localhost:9315/mcp with a bearer token.

Verify the setup

# Check server status
navra status

# Query the audit log
navra audit --limit 5

# List available tools
navra status  # shows tool count and registered agents

Next steps

  • Integrations — detailed guides for Claude Code, Goose, OpenAI clients, LangGraph
  • Guides — multi-agent flows, personas, RAG, memory, model server, agent bundles
  • Configuration — full config.toml reference
  • CLI Reference — all commands and flags
  • Architecture — microkernel design, crate layering, security model