# io.usefulapi/together-ai (remote · together-ai.usefulapi.io)

Run Together AI chat, embeddings and images; manage fine-tunes, batches and endpoints.

- Trust score: 78/100 (medium)
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-10-04

## Components

- remote · `together-ai.usefulapi.io`: 78/100 (this document), [markdown](https://verifymcp.io/servers/io-usefulapi-together-ai/together-ai.md), [page](https://verifymcp.io/servers/io-usefulapi-together-ai/together-ai)

## Channel facts

- Endpoint: `https://together-ai.usefulapi.io/mcp`
- Transports: `streamable-http`
- Auth: `none`
- Version: `1.0.0`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically against the live server, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-10-04.

- **Endpoint Security**: 89/100
  - The endpoint's TLS certificate is valid, in date, and uses a strong key.
  - Authorisation is enforced on tool calls, advertised via RFC 9728 protected-resource metadata. Discovery is public, which costs nothing: no tool can be invoked without a token.
  - HTTPS is enforced; there's no plaintext access path.
  - HSTS check failed: the Strict-Transport-Security header is absent.
  - DNSSEC check failed: this domain isn't protected by DNSSEC.
  - The authorisation server offers only Dynamic Client Registration (RFC 7591), which MCP 2026-07-28 deprecated in favour of Client ID Metadata Documents.
- **Transport & Reachability**: 100/100
  - Verified streamable-http transport via a live MCP handshake.
- **Schema Quality & AI Usability**: 81/100
  - AI-judged instruction clarity (excellent).
  - Tool/resource definitions use about 2082 tokens (~90/item across 23 items; 23 tools + 0 resources), lean.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 13/100
  - Stability observed for 4 of 30 days with no destabilising changes; credit accrues until the full window elapses.
- **Tool Coverage**: 95/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 85% of tool parameters carry a description.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - All 1 tool(s) whose name or description implies an irreversible operation declare an MCP destructiveHint annotation.
  - An AI judge read all 23 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 60/100
  - Spec-recency check failed: implements MCP spec 2025-06-18; the latest is 2026-07-28.

## Install

### How do I install the io.usefulapi/together-ai MCP server?

io.usefulapi/together-ai is a hosted endpoint at https://together-ai.usefulapi.io/mcp, so there is nothing to install locally. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add --transport http io-usefulapi-together-ai 'https://together-ai.usefulapi.io/mcp'
```

### Cursor

```json
{
  "mcpServers": {
    "io-usefulapi-together-ai": {
      "url": "https://together-ai.usefulapi.io/mcp"
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "io-usefulapi-together-ai": {
      "type": "http",
      "url": "https://together-ai.usefulapi.io/mcp"
    }
  }
}
```

### Codex

```toml
[mcp_servers.io-usefulapi-together-ai]
url = "https://together-ai.usefulapi.io/mcp"
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "io-usefulapi-together-ai": {
      "type": "remote",
      "url": "https://together-ai.usefulapi.io/mcp",
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add io-usefulapi-together-ai --url 'https://together-ai.usefulapi.io/mcp' --transport streamable-http
```

### Hermes

```yaml
mcp_servers:
  io-usefulapi-together-ai:
    url: "https://together-ai.usefulapi.io/mcp"
```

### Netclaw

```json
{
  "McpServers": {
    "io-usefulapi-together-ai": {
      "Transport": "http",
      "Url": "https://together-ai.usefulapi.io/mcp"
    }
  }
}
```

### Vellum

```bash
assistant mcp add io-usefulapi-together-ai -t streamable-http -u 'https://together-ai.usefulapi.io/mcp'
```

### Other

```json
{
  "mcpServers": {
    "io-usefulapi-together-ai": {
      "type": "http",
      "url": "https://together-ai.usefulapi.io/mcp"
    }
  }
}
```

The mcpServers block is a cross-client convention. Remote transports vary, so check your client's docs.

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-10-03 (score 78, +1)

No change was recorded against any check on this day. Stability & Change Management went from 7 to 10. That category is still filling its 30-day observation window: 2 days of observed history at the previous scan, 3 at this one. The score rises as the window fills, whether or not the server changes.

### 2026-10-02 (score 77, 0)

- [functional] Server version: 1.3.0 → 1.5.1

### 2026-10-01 (score 77, +5)

- [security improvement] HTTPS: unverified → pass
- [functional improvement] Stability: unverified → 0.03
- [functional] Server version: 1.0.0 → 1.3.0

### 2026-09-30 (score 72, +54)

- [security improvement] Authorization: unverified → pass
- [security improvement] Injection markers: unverified → pass
- [security improvement] Transport: fail → pass
- [security] First check of Judged manipulation: pass
- [security] First check of Authorization: partial
- [functional regression] MCP protocol: unverified → fail
- [functional improvement] Tool coverage: unverified → 100
- [functional] First check of Schema quality: fail
- [functional] First check of Schema quality: excellent
- [functional] First check of Tool coverage: 85
- [functional] First check of Destructive annotations: 100
- [functional] First check of Schema quality: pass

### 2026-09-29 (score 18)

First indexed and scored.

## MCP tools (23)

### `together_whoami` (~59 tokens)

Who am I

Identify the API key: its organization, project and project slug. The project slug forms the `<project_slug>/<endpoint_slug>` model name for dedicated-endpoint inference. A cheap way to confirm the key works. Together: GET /whoami.

### `together_list_models` (~64 tokens)

List models

List Together's models with type (chat, language, code, image, embedding, moderation, rerank), context length, organization, license and per-token pricing. Together: GET /models.

Input parameters:

- `dedicated` (boolean): Only return models that can run on dedicated endpoints.

### `together_list_files` (~44 tokens)

List files

List uploaded data files (fine-tune, eval and batch-api inputs, plus job outputs) with size, type, purpose and validation status. Together: GET /files.

### `together_get_file` (~60 tokens)

Get one file

Fetch one file's metadata, including its processing_status and validation_report (why a fine-tune training file was rejected). Together: GET /files/{id}.

Input parameters:

- `file_id` (string, required): The file id, e.g. file-abc123.

### `together_list_fine_tunes` (~38 tokens)

List fine-tuning jobs

List fine-tuning jobs with status, base model, output model name and training settings. Together: GET /fine-tunes.

### `together_get_fine_tune` (~67 tokens)

Get one fine-tuning job

Fetch one fine-tuning job: status, progress, hyperparameters, token counts, cost and the output model name. Together: GET /fine-tunes/{id}.

Input parameters:

- `fine_tune_id` (string, required): The job id, e.g. ft-abc123.

### `together_list_fine_tune_events` (~79 tokens)

List a fine-tuning job's events

List the event log of one fine-tuning job (queued, started, checkpoint saved, epoch completed, errors). The first place to look when a job failed. Together: GET /fine-tunes/{id}/events.

Input parameters:

- `fine_tune_id` (string, required): The job id, e.g. ft-abc123.

### `together_list_batches` (~33 tokens)

List batch jobs

List batch inference jobs with status, progress, model and input/output/error file ids. Together: GET /batches.

### `together_get_batch` (~68 tokens)

Get one batch job

Fetch one batch job: status (VALIDATING, IN_PROGRESS, COMPLETED, FAILED, EXPIRED, CANCELLED), progress, and the output_file_id / error_file_id once done. Together: GET /batches/{id}.

Input parameters:

- `batch_id` (string, required): The batch job id.

### `together_list_endpoints` (~101 tokens)

List endpoints

List endpoints with model, owner and state (PENDING, STARTING, STARTED, STOPPING, STOPPED, ERROR). Use mine=true and type=dedicated to see what is running on your account and billing by the minute. Together: GET /endpoints.

Input parameters:

- `mine` (boolean): Only endpoints owned by the caller.
- `type` (string): Filter by endpoint type.
- `usage_type` (string): Filter by usage type.

### `together_get_endpoint` (~58 tokens)

Get one endpoint

Fetch one dedicated endpoint: state, model, hardware, autoscaling bounds and display name. Together: GET /endpoints/{endpointId}.

Input parameters:

- `endpoint_id` (string, required): The endpoint id, e.g. endpoint-d23901de-....

### `together_list_hardware` (~63 tokens)

List hardware

List hardware configurations for dedicated endpoints with GPU type/count/memory and price in cents per minute. Pass a model to get only compatible configurations with live availability. Together: GET /hardware.

Input parameters:

- `model` (string): Only hardware compatible with this model, with availability.

### `together_list_evaluations` (~79 tokens)

List evaluation jobs

List LLM-as-a-judge evaluation jobs (classify, score, compare) with status, parameters and results once completed. Together: GET /evaluation.

Input parameters:

- `limit` (integer): Maximum number of jobs to return.
- `status` (string): Filter by status: pending, queued, running, completed, error, user_error.

### `together_chat_completion` (~177 tokens)

Chat completion

Run a chat completion on a Together model (billed per token). Non-streaming. For a dedicated endpoint pass its `<project_slug>/<endpoint_slug>` as the model. Together: POST /chat/completions.

Input parameters:

- `max_tokens` (integer): Maximum tokens to generate.
- `messages` (array, required): The conversation so far.
- `model` (string, required): Model name, e.g. meta-llama/Llama-3.3-70B-Instruct-Turbo.
- `reasoning_effort` (string): Reasoning effort for reasoning models that support it.
- `repetition_penalty` (number)
- `seed` (integer): Seed for reproducible sampling.
- `stop` (array): Stop sequences.
- `temperature` (number)
- `top_k` (integer)
- `top_p` (number)

### `together_create_embeddings` (~71 tokens)

Create embeddings

Generate vector embeddings for one or more texts (billed per token). Together: POST /embeddings.

Input parameters:

- `input` (required): A text, or a list of texts, to embed.
- `model` (string, required): Embedding model, e.g. BAAI/bge-large-en-v1.5.

### `together_generate_image` (~208 tokens)

Generate an image

Generate images from a prompt (billed per image/megapixel). Returns image URLs by default rather than base64, to keep responses small. Together: POST /images/generations.

Input parameters:

- `guidance_scale` (number): Prompt adherence; higher is more literal.
- `height` (integer): Height in pixels.
- `image_url` (string): Input image URL, for models that support editing.
- `model` (string, required): Image model, e.g. black-forest-labs/FLUX.1-schnell.
- `n` (integer): Number of images.
- `negative_prompt` (string): What to steer away from.
- `output_format` (string)
- `prompt` (string, required): What to draw.
- `response_format` (string): url (default here) or base64. base64 can be very large.
- `seed` (integer)
- `steps` (integer): Number of generation steps.
- `width` (integer): Width in pixels.

### `together_create_fine_tune` (~277 tokens)

Create a fine-tuning job

Start a fine-tuning job on an uploaded training file (billed per token processed). Stop it with together_cancel_fine_tune. Together: POST /fine-tunes.

Input parameters:

- `batch_size`: Batch size, or 'max' (the default).
- `from_checkpoint` (string): Continue from a previous job: <job_id>, <output_model_name>, optionally with :<step>.
- `learning_rate` (number)
- `max_seq_length` (integer)
- `model` (string, required): Base model to fine-tune.
- `n_checkpoints` (integer): Intermediate checkpoints to save.
- `n_epochs` (integer): Passes over the training data.
- `n_evals` (integer): Evaluations on the validation set during training.
- `suffix` (string): Suffix for the fine-tuned model's name (max 64 chars).
- `training_file` (string, required): File id of an uploaded training file (purpose fine-tune).
- `training_method`: Supervised fine-tuning (sft, the default) or preference tuning (dpo).
- `training_type`: Full fine-tune or LoRA. Together defaults to LoRA when omitted.
- `validation_file` (string): File id of an uploaded validation file.
- `warmup_ratio` (number)

### `together_cancel_fine_tune` (~71 tokens)

Cancel a fine-tuning job

Cancel a running fine-tuning job. Cannot be resumed, but a new job can continue from its last checkpoint via from_checkpoint. Together: POST /fine-tunes/{id}/cancel.

Input parameters:

- `fine_tune_id` (string, required): The job id, e.g. ft-abc123.

### `together_create_batch` (~119 tokens)

Create a batch job

Start an asynchronous batch job over an uploaded JSONL input file (purpose batch-api), at a discount to real-time inference. Together: POST /batches.

Input parameters:

- `completion_window` (string): Time window for completion, e.g. 24h.
- `endpoint` (string, required): The API each line of the input file is sent to.
- `input_file_id` (string, required): File id of the uploaded JSONL request file.
- `model_id` (string): Model to process the requests with.
- `priority` (integer): Processing priority.

### `together_cancel_batch` (~41 tokens)

Cancel a batch job

Cancel a batch job that has not finished. Together: POST /batches/{id}/cancel.

Input parameters:

- `batch_id` (string, required): The batch job id.

### `together_create_endpoint` (~188 tokens)

Create a dedicated endpoint

Deploy a model on dedicated GPUs. The endpoint STARTS AUTOMATICALLY and bills per minute of uptime until stopped — set inactive_timeout to auto-stop it, and use together_list_hardware for valid hardware ids. Together: POST /endpoints.

Input parameters:

- `autoscaling` (object, required): Replica bounds for autoscaling.
- `availability_zone` (string): Availability zone, e.g. us-central-4b.
- `disable_speculative_decoding` (boolean)
- `display_name` (string): Human-readable name.
- `hardware` (string, required): Hardware id, e.g. 1x_nvidia_a100_80gb_sxm.
- `inactive_timeout` (integer): Minutes of inactivity before auto-stop; 0 disables it.
- `model` (string, required): The model to deploy.
- `state` (string): Initial state. Pass STOPPED to create without starting (and without billing).

### `together_start_endpoint` (~58 tokens)

Start a dedicated endpoint

Start a stopped dedicated endpoint. It bills per minute of uptime until stopped. Reversible with together_stop_endpoint. Together: PATCH /endpoints/{endpointId} with state=STARTED.

Input parameters:

- `endpoint_id` (string, required): The endpoint id.

### `together_stop_endpoint` (~59 tokens)

Stop a dedicated endpoint

Stop a running dedicated endpoint, which stops its per-minute billing. Requests to it fail until it is started again. Together: PATCH /endpoints/{endpointId} with state=STOPPED.

Input parameters:

- `endpoint_id` (string, required): The endpoint id.

## Diagnostics

Captured diagnostic sections: TLS, DNSSEC, Authorisation, Transports. The full working is on the page: https://verifymcp.io/servers/io-usefulapi-together-ai/together-ai#diagnostics

## Score history

- 2026-10-04: 78
- 2026-10-03: 78
- 2026-10-02: 77
- 2026-10-01: 77
- 2026-09-30: 72
- 2026-09-29: 18

## Common questions

### What is the io.usefulapi/together-ai MCP server?

io.usefulapi/together-ai is an MCP server listed in the public MCP registry as io.usefulapi/together-ai. Run Together AI chat, embeddings and images; manage fine-tunes, batches and endpoints. This page covers its hosted endpoint (https://together-ai.usefulapi.io/mcp).

### Is the io.usefulapi/together-ai MCP server safe to use?

io.usefulapi/together-ai scores 78 out of 100 on VerifyMCP. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the io.usefulapi/together-ai MCP server expose?

io.usefulapi/together-ai exposes 23 tools: together_whoami, together_list_models, together_list_files, together_get_file, together_list_fine_tunes, and 18 more. Their descriptions and schemas cost roughly 2,082 tokens of context every time the server is loaded.

### Does the io.usefulapi/together-ai MCP server require authentication?

Yes. io.usefulapi/together-ai asked us for credentials when we connected, so you will need to authorise it in your MCP client before it can do anything.

### Is the io.usefulapi/together-ai MCP server still maintained?

io.usefulapi/together-ai is still listed as active in the MCP registry. We last reached this channel on 4 October 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

## Links

- Remote endpoint: https://together-ai.usefulapi.io/mcp
- Repository: https://github.com/m190/usefulapi-mcp
- Changelog RSS feed: https://verifymcp.io/servers/io-usefulapi-together-ai/together-ai.xml
- Changelog JSON feed: https://verifymcp.io/servers/io-usefulapi-together-ai/together-ai.json
- HTML version of this page: https://verifymcp.io/servers/io-usefulapi-together-ai/together-ai