Local LLM Setup

Use Keplars MCP tools with locally running models via Ollama, AnythingLLM, Continue, Jan, or LibreChat.

Local LLMs can use the Keplars MCP server, but you need two things:

A tool-calling capable model — the model must support function/tool calling. Not all local models do this reliably.
An MCP-capable client — the local model does not speak MCP on its own. A client wrapper bridges the model to MCP tools.

Model quality matters

MCP tool use requires the model to correctly decide when to call a tool, which arguments to pass, and how to interpret the result. Smaller or older models often fail at this. Use a model with at least 8B parameters that explicitly supports tool calling.

Recommended models (via Ollama):

Model	Pull command
Llama 3.1 8B	`ollama pull llama3.1`
Qwen2.5 7B	`ollama pull qwen2.5`
Mistral 7B	`ollama pull mistral`
Llama 3.3 70B	`ollama pull llama3.3` (if you have the VRAM)

Client Setup

AnythingLLM has built-in MCP support and works with Ollama out of the box.

Point AnythingLLM at your Ollama instance

In AnythingLLM Settings > LLM Provider, select Ollama and set the base URL to http://localhost:11434. Choose a tool-calling capable model.

Open MCP settings

Go to Settings > Agent > MCP Servers and click the + button to add a new server.

Add the Keplars server

AnythingLLM connects to remote MCP servers directly over HTTP. Paste the following into the JSON config editor:

{
  "mcpServers": {
    "keplars": {
      "url": "https://mcp.keplars.com/mcp",
      "type": "streamable",
      "disabled": false,
      "alwaysAllow": [],
      "headers": {
        "Authorization": "<YOUR_SEND_KEY>",
        "X-Admin-Api-Key": "<YOUR_ADMIN_KEY>"
      }
    }
  }
}

Enable the server and test

Toggle the Keplars server on and start a chat. Ask your model to send a test email to verify tools are reachable.

Continue is a VS Code and JetBrains extension that supports MCP with any Ollama model.

Install Continue

Install the Continue extension from the VS Code marketplace.

Open the Continue config

Press Cmd+Shift+P (macOS) or Ctrl+Shift+P (Windows) and run Continue: Open config.yaml.

Add Ollama as the model provider

models:
  - name: llama3.1
    provider: ollama
    model: llama3.1

Add the Keplars MCP server

Continue supports direct HTTP MCP connections — no mcp-remote needed:

mcpServers:
  - name: keplars
    type: streamable-http
    url: https://mcp.keplars.com/mcp
    requestOptions:
      headers:
        Authorization: "<YOUR_SEND_KEY>"
        X-Admin-Api-Key: "<YOUR_ADMIN_KEY>"

Reload Continue

Save the config file. Continue reloads automatically. The Keplars tools appear in the tools panel.

Jan is a local-first AI desktop app with native MCP support.

Download and install Jan

Download Jan from jan.ai and install a tool-calling model from the model hub (Llama 3.1 or Qwen2.5 recommended).

Open MCP settings

Go to Settings > MCP Servers.

Install mcp-remote

Jan uses stdio to bridge local MCP servers. Install it once:

npm install -g mcp-remote

Add the Keplars server

Click the + button and paste the following config:

{
  "mcpServers": {
    "keplars": {
      "command": "mcp-remote",
      "args": [
        "https://mcp.keplars.com/mcp",
        "--header",
        "Authorization:<YOUR_SEND_KEY>",
        "--header",
        "X-Admin-Api-Key:<YOUR_ADMIN_KEY>"
      ]
    }
  }
}

Restart Jan

Quit and reopen Jan. The Keplars tools will be available in chat.

LibreChat is a self-hosted chat UI that supports Ollama and MCP servers.

Install mcp-remote

LibreChat uses stdio to bridge to the Keplars MCP endpoint. Install it on the host running LibreChat:

npm install -g mcp-remote

Configure Ollama in LibreChat

In your LibreChat librechat.yaml, add Ollama as an endpoint and set a tool-calling model as default.

Add the MCP server

In librechat.yaml under mcpServers:

mcpServers:
  keplars:
    command: mcp-remote
    args:
      - https://mcp.keplars.com/mcp
      - --header
      - "Authorization:<YOUR_SEND_KEY>"
      - --header
      - "X-Admin-Api-Key:<YOUR_ADMIN_KEY>"

Restart LibreChat

docker compose restart

The Keplars tools will be available when using Ollama models in any conversation.

Troubleshooting

Model ignores tools or hallucinates tool calls

Switch to a larger or newer model. Llama 3.1 8B and Qwen2.5 7B have the most reliable tool-calling behaviour among commonly available local models.

mcp-remote not found

Run npm install -g mcp-remote and make sure the npm global bin is in your PATH.

Tools connect but email sends fail with "Unauthorized"

Check that your send key (live_ prefix) is correctly placed in the Authorization arg with no space after the colon: Authorization:<key> not Authorization: <key>.

Tools connect but template/contact/campaign tools fail

These tools require the admin key in X-Admin-Api-Key. Make sure both headers are present in your config.

Local LLM Setup

On this page