Bifrost is an AI-native API gateway — purpose-built for routing traffic to large language models and AI services rather than traditional REST backends. Where conventional gateways manage developer portals and product catalogs, Bifrost manages model endpoints, chat routing, and the health of an AI service layer. The Naftiko capability for Bifrost creates a governed management surface for the infrastructure that routes AI traffic.
What the Capability Covers
The ai-chat-routing capability wraps the Bifrost management API and exposes three core operations: chat completion routing, streaming chat, and gateway health. Bifrost supports 20+ AI providers with automatic failover and load balancing.
MCP Tools Available
create-chat-completion— Route a chat prompt through Bifrost to any configured AI provider usingprovider/model-nameformatstream-chat-completion— Stream a chat completion response for real-time output workflowsget-gateway-health— Check the health of the Bifrost gateway and all configured providers
Capability Specification
naftiko: "1.0.0-alpha1"
info:
label: AI Chat Routing
description: >-
Workflow capability for routing AI chat completions through the Bifrost
gateway to 20+ AI providers with automatic failover and load balancing.
tags:
- Bifrost
- AI Gateway
- LLM
- Chat Routing
binds:
- namespace: env
keys:
BIFROST_API_KEY: BIFROST_API_KEY
BIFROST_BASE_URL: BIFROST_BASE_URL
capability:
consumes:
- namespace: bifrost
type: http
baseUri: ""
authentication:
type: bearer
token: ""
resources:
- name: chat-completions
path: "/v1/chat/completions"
operations:
- name: create-chat-completion
method: POST
- name: stream-chat-completion
method: POST
- name: health
path: "/health"
operations:
- name: get-health
method: GET
exposes:
- type: mcp
port: 9080
namespace: bifrost-routing-mcp
transport: http
description: MCP server for AI-assisted access to the Bifrost AI gateway.
tools:
- name: create-chat-completion
description: >-
Send a chat message to any AI provider using the provider/model-name
format. Supports OpenAI, Anthropic, Cohere, and 20+ other providers.
hints:
readOnly: false
destructive: false
idempotent: false
call: bifrost.create-chat-completion
outputParameters:
- type: object
mapping: "$."
- name: stream-chat-completion
description: >-
Stream a chat completion response as server-sent events from any
supported AI provider.
hints:
readOnly: false
destructive: false
idempotent: false
call: bifrost.stream-chat-completion
outputParameters:
- type: object
mapping: "$."
- name: get-gateway-health
description: Check the health of the Bifrost gateway and all configured providers.
hints:
readOnly: true
openWorld: true
call: bifrost.get-health
outputParameters:
- type: object
mapping: "$."
- type: rest
port: 8080
namespace: bifrost-routing-rest
Running It
docker pull ghcr.io/naftiko/framework:latest
# Set BIFROST_API_KEY and BIFROST_BASE_URL in your .env file
docker run -p 9080:9080 \
-v ./capabilities/ai-chat-routing.yaml:/app/capability.yaml \
--env-file .env \
ghcr.io/naftiko/framework:latest /app/capability.yaml
Using Naftiko capabilities in front of Bifrost creates a layered AI infrastructure: Bifrost routes AI traffic with intelligent load balancing across model providers, while Naftiko governs access to that routing layer with a specification-driven integration. The Bifrost capability is part of the Naftiko Fleet.
