Use Qoder with Eden AI

Qoder accepts custom OpenAI-compatible and Anthropic-compatible model providers, so Eden AI needs no plugin: you enter a base URL, a key and a model ID in the desktop app, and the model appears in the chat’s model picker and in the Qoder CLI. Qoder sends tool definitions with every request, so pick a model that supports function calling.

Overview

Qoder is an AI coding agent that comes as a desktop app and a CLI. Eden AI plugs into it in two ways:

As a custom model provider: one API key gives the agent 500+ models from OpenAI, Anthropic, Google, Mistral and more, with unified billing and monitoring.
As an MCP server: Eden AI’s expert models (OCR, web search, speech, translation and more) become tools the agent can call.

This page was tested with the Qoder desktop app and Qoder CLI v1.1.83.

Prerequisites

The Qoder desktop app installed
Eden AI API key from app.edenai.run → API Keys

Setup

1. Add Eden AI as a model provider

Open Settings (the gear icon, bottom left), go to Models and click Add model.

The Models page in Qoder settings, listing custom models with an Add model button

In the Add model dialog, fill in the fields:

Field	Value
Provider	OpenAI Compatible
API type	Chat Completions API (recommended)
Base URL	`https://api.edenai.run/v3`
API key	Your Eden AI API key
Model IDs	e.g. `claude-sonnet-5.5` or `gpt-5.5`

A Model ID is a canonical model name. It names the model only, so Eden AI chooses the provider that serves it, by price, speed or latency, with automatic failover. The full list of canonical names is at GET https://api.edenai.run/v3/models?view=models (public, no key needed). Click + Add Model ID to add more models on the same endpoint, then click Next.

The Add model dialog with OpenAI Compatible, Chat Completions API, the Eden AI base URL, an API key and one model ID

2. Configure model capabilities

The Configure model capabilities screen describes the model to Qoder:

Display name: the name shown in the model picker.
Supported context windows and Default context window: use the model’s context_length from GET https://api.edenai.run/v3/models?view=models (see List Models).
Capabilities: turn on Vision and Thinking when the model supports them.
Reasoning effort: Minimal, Low, Medium, High, Extra high or Maximum. See Reasoning effort for what Qoder sends for each.

The Configure model capabilities screen with display name, context windows, Vision and Thinking capabilities and reasoning effort buttons

Click Validate and save. The Models saved dialog confirms the model was added; click Done.

The Models saved dialog confirming the model configuration check passed and one model was added

Validate and save does not test the connection. A wrong key is only reported when you send the first message; see Troubleshooting.

Provider, API type and model cannot be changed after adding. To switch, add a new model.

3. Pick the model in the chat

Open the model picker at the bottom of the chat and choose the model under Custom. The picker also lets you change the Thinking effort of the selected model.

The chat model picker on the Custom tab, listing the Eden AI models and the Thinking effort of the selected one

Other API types

Chat Completions API is the recommended API type. The two others work with the same Base URL:

OpenAI Compatible with Responses API: Qoder sends the full conversation history each turn, so no server-side state is needed.
Anthropic Compatible: Qoder calls /v1/messages under the same Base URL, https://api.edenai.run/v3. Use it for Claude models only: in this mode, a non-Claude model (Devstral) stopped without using its tools, while the same model worked with Chat Completions.

Model naming

Qoder sends the Model ID as you write it, so use a canonical model name as listed by GET https://api.edenai.run/v3/models?view=models, such as claude-sonnet-5.5 or gpt-5.5. Qoder sends tool definitions with every request, so choose a model whose entry has supports_function_calling: true. Claude Sonnet, GPT, Gemini Flash and Devstral were tested end to end with Chat Completions. List the available models, with their capabilities and pricing, in List Models.

Reasoning effort

The effort chosen in the model settings or the chat picker is sent as the parameter of the API type in use:

API type	What Qoder sends
Chat Completions API	`reasoning_effort`: `low`, `medium`, `high`, `xhigh` or `max`. Minimal sends nothing.
Responses API	`reasoning.effort`
Anthropic Compatible	`thinking` in adaptive mode, plus `effort`

Switching models

Open the model picker in the chat and choose another model; switching mid-chat works. To add a model, or to change the provider or API type of one, go to Settings > Models > Add model again.

Costs

Prompt caching works without any setup: Qoder sends a session id header per chat, so the agent’s steps reuse 30,000 to 50,000 cached tokens. See Prompt Caching.
With Auto approve on, Qoder asks the selected model for a short safety check before each tool call. These small requests also go through Eden AI.

Qoder CLI

The CLI uses the same custom models as the desktop app. Qoder’s documentation lists custom models under the Individual plan; this was tested on a Pro trial.

Install

On Windows, run in PowerShell:

irm https://qoder.com/install.ps1 | iex

The command is qodercli. If qodercli is not recognized, the binary is in %USERPROFILE%\.qoder\bin\qodercli\: add that folder to your PATH and open a new terminal.

Pick a model

Run qodercli in your project folder, then /model and open the Custom tab. Models added in the desktop app appear automatically under Local endpoints on the same computer.

The Custom tab of /model in the Qoder CLI, listing the models added in the desktop app under Local endpoints with their effort

To add a model from the CLI instead, choose OpenAI-compatible… (Chat Completions / Responses) or Anthropic-compatible… under Custom providers and enter the same values as in the desktop app.

The Custom providers list in the Qoder CLI: OpenAI-compatible and Anthropic-compatible, above the platform presets

Keys in the model list: ←/→ change the effort, e edits a model, n adds one, d deletes one and s selects it for this session only.

Expert Models as Tools (MCP)

Eden AI also hosts an MCP server that exposes the expert model catalog (OCR, web search, speech, translation and more) as tools. In Qoder, open Extensions in the left sidebar, go to Connectors, click Add Connector and choose Add custom MCP. On the Form tab, fill in:

Field	Value
MCP name	`edenai`
Transport	HTTP
Server URL	`https://mcp.edenai.run/mcp`
Headers	`Authorization` = `Bearer YOUR_EDEN_AI_API_KEY`

Click Add MCP.

The Add custom MCP form with the name edenai, HTTP transport, the Eden AI MCP server URL and an Authorization header

The JSON tab takes the same server as a block. Qoder saves it under mcpServers in ~/.qoder/settings.json, which the CLI reads too:

{
  "mcpServers": {
    "edenai": {
      "type": "streamable-http",
      "url": "https://mcp.edenai.run/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_EDEN_AI_API_KEY"
      }
    }
  }
}

The connector shows Ready and lists its 42 tools; the switch next to it turns it on and off. Qoder loads the tools on demand, so they do not add to every request.

The edenai connector marked Ready under Custom Connectors, with its tools listed

Try it with a prompt such as: “Use the Eden AI web search tool to find the latest stable Python release.”

A Qoder chat where the agent ran the Eden AI web search tool and answered with the latest Python release and its sources

A few things to know about the tools:

Files: pass public URLs. Some sites block automated downloads and answer 403. For a local file, upload it through the file upload API first and give the agent the file ID in a code block. Do not ask the agent to upload a large file itself: it has to write the file out as base64 and hits its output limit.
Billing: each tool call is billed as a normal expert model call and shows in monitoring.

See the MCP server page for the full tool catalog and the conventions its tools share.

Troubleshooting

Custom model authentication failed

The message reads “Authentication for the custom model failed. Please check whether the API key and related configuration are correct.” The key is wrong or expired: copy it again from app.edenai.run under API Keys, with no leading or trailing spaces, and check the model’s configuration under Settings > Models.

Response incomplete

“Response incomplete. The model reached its output token limit” means the answer did not fit in the model’s output. Narrow the request. It is typical when the agent tries to pass large file content through a tool; see Files above.

MCP tool not found

A message ending in “not found” for an MCP tool is transient. Retry; the agent usually recovers on its own.

Warning about HTTP addresses

A yellow “HTTP addresses can be used…” warning in the model settings means the Base URL does not start with https://. Use https://api.edenai.run/v3.

Next Steps

Cline: VS Code coding agent on Eden AI
Claude Code: Anthropic’s official CLI coding agent on Eden AI
List Models: browse the catalog with capabilities and pricing
MCP server: the full expert model tool catalog

V3 Documentation

Quick Start

Overview

LLMs

Expert Models

General

Organization

Data Governance

Integrations

Overview

Prerequisites