Coding agent setup

Together AI publishes two complementary tools for coding agents:

Skill: A single together-ai skill that teaches your agent how to write correct Together AI code (right model IDs, SDK patterns, best practices) across every product area.
Docs MCP server: Gives your agent live access to this documentation site so it can look up current information without leaving your editor.

Install both for the best experience: the skill for code generation and MCP for documentation lookup.

To run the coding agent itself on Together AI models, use Together Link. It routes coding agents through Together AI with one command and no provider configuration. The together-ai skill also covers Together Link setup.

Skill

The together-ai skill covers the whole platform, from chat completions and image generation to fine-tuning, dedicated model inference, and GPU clusters. When your agent detects a Together AI task, it loads the skill automatically. You can also invoke it by name, for example /together-ai in Claude Code or $together-ai in Codex. Installing one skill for the whole platform doesn’t fill your agent’s context. The skill’s short SKILL.md routes each task to a guide for that product area, and the agent opens the deeper model lists, API references, and runnable scripts only when that guide points to them.

Install the skill

npx skills add togethercomputer/skills

# From the plugin marketplace
/plugin marketplace add togethercomputer/skills
/plugin install togetherai-skills@togetherai-skills

# Or copy manually (project-level)
cp -r skills/together-ai your-project/.claude/skills/

# Or copy manually (global, available in all projects)
cp -r skills/together-ai ~/.claude/skills/

cp -r skills/together-ai your-project/.cursor/skills/

cp -r skills/together-ai your-project/.agents/skills/

gemini extensions install https://github.com/togethercomputer/skills.git --consent

To verify a manual install, check that the skill’s SKILL.md is in place (for example, ls your-project/.claude/skills/together-ai/SKILL.md).

The skill replaces the earlier per-product skills (together-chat-completions, together-images, together-gpu-clusters, and so on). If you installed any of them, delete their directories after you install together-ai, so your agent doesn’t load outdated guidance alongside the new skill.

What the skill covers

Each product area has its own domain guide inside the skill.

Domain guide	What it covers
Chat completions	Serverless chat inference, streaming, multi-turn conversations, tool calling, structured JSON outputs, and reasoning and vision models.
Images	Text-to-image generation, image editing with Kontext, FLUX model selection, and reference-image guidance.
Video	Text-to-video and image-to-video generation, keyframe control, and async job polling.
Audio	Text-to-speech (REST, streaming, realtime WebSocket) and speech-to-text (transcription, translation, diarization, timestamps, realtime).
Batch inference	Async batch jobs with JSONL input, polling, result downloads, and up to 50% cost savings.
Evaluations	LLM-as-a-judge jobs that classify, score, or compare model outputs.
Fine-tuning	Low-rank adapter (LoRA), full, DPO preference, VLM, function-calling, and reasoning fine-tuning, plus bring-your-own-model uploads.
Sandboxes	Remote stateful Python execution for agent-written code, data analysis, and charts.
Dedicated model inference	Endpoints, deployments, and configs on dedicated GPUs, with autoscaling, traffic splitting, A/B tests, shadow experiments, metrics, custom model or LoRA uploads, and migration from the v1 endpoints API.
Dedicated container inference	Custom Dockerized inference workers using the Jig CLI, Sprocket SDK, and queue API.
GPU clusters	On-demand and reserved GPU clusters with Kubernetes or Slurm, shared storage, and credentials.
Kueue	Quota-gated job queueing on a Together AI Kubernetes cluster.
Volcano	All-or-nothing gang scheduling on a Together AI Kubernetes cluster.
Together Link	Running Claude Code, Codex, OpenCode, Pi Code, Claude Desktop, or ChatGPT Desktop on Together AI models.

Single-area tasks

Describe what you want, and the skill routes the agent to the matching domain guide. If you prompt your agent with:

Build a multi-turn chatbot using Together AI with Kimi-K3
that can call a weather API and return structured JSON.

The agent opens the chat completions guide and generates correct SDK code with the right model ID, streaming setup, tool definitions, and the complete tool-call loop.

Multi-area tasks

For tasks that span several products, the agent reads each domain guide when it reaches that step. If you prompt your agent with:

Fine-tune a model on my support tickets with Together AI, deploy it
to a dedicated endpoint, then evaluate its answers with an LLM judge.

The agent works through three guides:

Fine-tuning: Validates and uploads the dataset, launches a LoRA job, and polls it to completion.
Dedicated model inference: Deploys the fine-tuned model and waits until the endpoint is ready.
Evaluations: Scores the endpoint’s answers with an LLM judge and downloads the per-row results.

See the skills repository for more workflow examples.

SDK compatibility

All generated code targets the Together Python v2 SDK (together>=2.0.0) and the Together TypeScript SDK (together-ai). If you’re upgrading from v1, see the Python v2 SDK migration guide.

Docs MCP server

Model Context Protocol (MCP) lets AI coding agents call external tools and pull in external data. The Together AI docs MCP server gives your agent direct access to this documentation site without leaving your editor.

Install

The fastest install is the universal npx add-mcp shortcut, which detects your active client and configures the server in one step. The other tabs cover client-specific install commands and manual configuration.

npx add-mcp https://docs.together.ai/mcp

claude mcp add --transport http "TogetherAIDocs" https://docs.together.ai/mcp

For manual configuration, add this to your Cursor MCP settings:

{
  "mcpServers": {
    "together-docs": {
      "url": "https://docs.together.ai/mcp"
    }
  }
}

Install in VS CodeFor manual configuration, add this to your VS Code settings.json:

{
  "mcp": {
    "servers": {
      "together-docs": {
        "type": "http",
        "url": "https://docs.together.ai/mcp"
      }
    }
  }
}

See the Codex repository for details. To connect to the remote server, add this to your Codex configuration:

[mcp_servers.together_docs]
type = "http"
url = "https://docs.together.ai/mcp"

Add this to your OpenCode configuration file:

{
  "mcp": {
    "together_docs": {
      "type": "remote",
      "url": "https://docs.together.ai/mcp",
      "enabled": true
    }
  }
}

Prompt examples

Once installed, your agent can answer prompts like:

“Write a script to process data with batch inference.”
“Build a simple chat app with Together AI’s chat completions API.”
“Find the best open-source model for frontier coding.”
“How do I fine-tune a model on my own data?”

The MCP server provides tools to search and retrieve documentation content, so your agent gets accurate answers without leaving your coding environment.

Skills repository on GitHub

Source, domain guides, references, and runnable scripts for the together-ai skill.

Together AI cookbook

End-to-end examples and tutorials.

Python v2 SDK migration guide

Breaking changes between the v1 and v2 SDKs.

Agent Skills specification

The open standard the skill follows.

GET STARTED

SERVERLESS

INFERENCE APIS

DEDICATED MODEL INFERENCE

DEDICATED CONTAINER INFERENCE

POST-TRAINING

FINE-TUNING

GPU CLUSTERS

CODE EXECUTION

ADMINISTRATION

Skill

Install the skill

What the skill covers

Single-area tasks

Multi-area tasks

SDK compatibility

Docs MCP server

Install

Prompt examples

Skills repository on GitHub

Together AI cookbook

Python v2 SDK migration guide

Agent Skills specification

​Skill

​Install the skill

​What the skill covers

​Single-area tasks

​Multi-area tasks

​SDK compatibility

​Docs MCP server

​Install

​Prompt examples

​Related resources

Skills repository on GitHub

Together AI cookbook

Python v2 SDK migration guide

Agent Skills specification

Skill

Install the skill

What the skill covers

Single-area tasks

Multi-area tasks

SDK compatibility

Docs MCP server

Install

Prompt examples

Related resources