Hashference

npx add-mcp 'https://hashference.com/mcp'

Connect your tool

Exact settings for the coding agents and apps people use with Hashference. Replace hf_... with your API key, and <model-id> with any ID from GET /v1/models.

Claude Code

Start it in one line:

Shell

ANTHROPIC_BASE_URL="https://hashference.com" ANTHROPIC_AUTH_TOKEN="hf_..." ANTHROPIC_API_KEY="" MAX_THINKING_TOKENS=0 ANTHROPIC_MODEL="<model-id>" ANTHROPIC_DEFAULT_HAIKU_MODEL="<model-id>" CLAUDE_CODE_SUBAGENT_MODEL="<model-id>" claude

To keep it set up:

Point Claude Code at Hashference with a few environment variables. This uses a low-cost open model for the main session, quick tasks and subagents. Anthropic compatibility does not imply the model is Claude.

Shell

export ANTHROPIC_BASE_URL="https://hashference.com"
export ANTHROPIC_AUTH_TOKEN="hf_..."
export ANTHROPIC_API_KEY=""
export MAX_THINKING_TOKENS=0
export ANTHROPIC_MODEL="<model-id>"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="<model-id>"
export CLAUDE_CODE_SUBAGENT_MODEL="<model-id>"
claude

ANTHROPIC_API_KEY must be empty, or Claude Code uses it instead. If you signed in to Claude Code before, run /logout first, then check /status. To keep the setup, put the same variables in the env block of ~/.claude/settings.json. For Claude itself, set ANTHROPIC_MODEL to a Claude Messages ID from the catalog and remove MAX_THINKING_TOKENS; a Claude request reserves up to about $1 of balance while it runs.

Cursor

In Cursor Settings, open Models. Under API Keys, paste your key as the OpenAI API key, turn on Override OpenAI Base URL, then add the model by name.

Settings

OpenAI API Key      hf_...
Override Base URL   https://hashference.com/v1
Custom model        <model-id>

Cursor calls the API from its own servers, so it needs this public HTTPS address rather than a local one.

VS Code

Install Claude Code for VS Code · Install Cline

Install the Claude Code extension or Cline, then give it these settings. The Claude Code extension reads ~/.claude/settings.json; Cline takes the fields in its settings panel.

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://hashference.com",
    "ANTHROPIC_AUTH_TOKEN": "hf_...",
    "ANTHROPIC_API_KEY": "",
    "MAX_THINKING_TOKENS": "0",
    "ANTHROPIC_MODEL": "<model-id>",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "<model-id>",
    "CLAUDE_CODE_SUBAGENT_MODEL": "<model-id>"
  }
}

Cline settings

API Provider   OpenAI Compatible
Base URL       https://hashference.com/v1
API Key        hf_...
Model ID       <model-id>

The install buttons open VS Code on the extension’s page. If you already have a settings.json, merge the env block into it. Roo Code and Kilo Code take the same fields as Cline.

Codex

Add Hashference as a model provider in ~/.codex/config.toml. Codex reads the key from the variable you name.

config.toml

model = "<model-id>"
model_provider = "quai"

[model_providers.quai]
name = "Hashference"
base_url = "https://hashference.com/v1"
env_key = "HASHFERENCE_API_KEY"
wire_api = "responses"

Shell

export HASHFERENCE_API_KEY="hf_..."
codex

Any model ID from GET /v1/models works as model.

OpenCode

Add a provider to opencode.json (in your project, or ~/.config/opencode/ for every project). The key stays in an environment variable.

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "quai": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Hashference",
      "options": {
        "baseURL": "https://hashference.com/v1",
        "apiKey": "{env:HASHFERENCE_API_KEY}"
      },
      "models": {
        "<model-id>": {
          "name": "<model-id>"
        }
      }
    }
  },
  "model": "quai/<model-id>"
}

Shell

export HASHFERENCE_API_KEY="hf_..."
opencode

Add more models under models with any ID from GET /v1/models, then pick them with /models.

Cline, Roo Code, Kilo Code

In the extension’s settings, choose OpenAI Compatible as the API provider and fill in these fields.

Settings

API Provider   OpenAI Compatible
Base URL       https://hashference.com/v1
API Key        hf_...
Model ID       <model-id>

The same fields work in Roo Code and Kilo Code.

Hermes Agent

Run hermes model and choose Custom endpoint, or set the model block in ~/.hermes/config.yaml and keep the key in ~/.hermes/.env.

config.yaml

model:
  provider: custom
  base_url: https://hashference.com/v1
  default: <model-id>
  key_env: HASHFERENCE_API_KEY
  context_length: 65536
  max_tokens: 8192

.env

HASHFERENCE_API_KEY=hf_...

Keep context_length and max_tokens as shown: Hermes otherwise assumes a larger context and output than the hosted model takes, and compresses too late. Requests over a model’s output limit are lowered to it. Run hermes doctor to check the connection.

OpenClaw

Merge this into ~/.openclaw/openclaw.json. The provider defines the model, and the agents entry allows it.

openclaw.json

{
  "models": {
    "mode": "merge",
    "providers": {
      "quai": {
        "baseUrl": "https://hashference.com/v1",
        "apiKey": "hf_...",
        "api": "openai-completions",
        "models": [
          {
            "id": "<model-id>",
            "name": "<model-id>",
            "input": [
              "text"
            ]
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "models": {
        "quai/<model-id>": {
          "alias": "quai"
        }
      }
    }
  }
}

Switch to it with /model quai, then check /status.

MCP server

Give any MCP-capable agent Hashference as tools: list models, check your balance, ask a model, and set up top-ups. Apps that support MCP sign-in (Claude, ChatGPT, Cursor, VS Code, Claude Code) need only the URL: you sign in, approve, and the app gets its own key. Others connect with your API key.

Claude or ChatGPT

Add a custom connector with this URL, then sign in and approve:
https://hashference.com/mcp

Claude: Settings > Connectors > Add custom connector.
ChatGPT: turn on developer mode in Settings, then add a connector with this URL.

Claude Code

claude mcp add --transport http hashference https://hashference.com/mcp \
  --header "Authorization: Bearer hf_..."

# Or sign in instead of pasting a key: leave out --header, then run /mcp in Claude Code.

Codex

# ~/.codex/config.toml
[mcp_servers.hashference]
url = "https://hashference.com/mcp"
bearer_token_env_var = "HASHFERENCE_API_KEY"

Cursor

{
  "mcpServers": {
    "hashference": {
      "url": "https://hashference.com/mcp",
      "headers": {
        "Authorization": "Bearer hf_..."
      }
    }
  }
}

The chat tool spends from your workspace balance, like any other request with this key. To run it locally instead, with a work-paid x402 tool, install the CLI (npm i -g https://hashference.com/hashference-0.1.4.tgz) and use: claude mcp add hashference -- hashference mcp

Any OpenAI-compatible app

Start it in one line:

Shell

export OPENAI_BASE_URL="https://hashference.com/v1" OPENAI_API_KEY="hf_..."

To keep it set up:

Most apps and SDKs read these two variables. Anthropic SDKs use ANTHROPIC_BASE_URL set to the origin instead.

Shell

export OPENAI_BASE_URL="https://hashference.com/v1"
export OPENAI_API_KEY="hf_..."

Use any ID from GET /v1/models as the model.