# Draft LinkedIn posts with a local Ollama model

Source: https://finchreach.com/blog/ollama-linkedin-local-drafts
Published 2026-10-28 by Finchreach team. Tags: ollama, linkedin, mcp, local-llm, content.

Ollama has no MCP client of its own. Here is the honest route: pair a local model with Goose, ollmcp or Open WebUI and Finchreach to draft LinkedIn posts.

## The short answer

Ollama can’t connect to LinkedIn on its own, because Ollama runs models and isn’t an MCP client. To use a local Ollama model with Finchreach, you put an MCP-capable client in between: Goose, ollmcp (mcp-client-for-ollama) or Open WebUI. The client uses your Ollama model for the thinking and Finchreach for the LinkedIn tools. A good first use case is drafting your LinkedIn posts from private notes with a model that runs on your own machine. Drafts go to your Finchreach [approval queue](https://finchreach.com/approvals) and only go live once you approve them.

MCP (Model Context Protocol) is a standard plug that lets AI apps use outside tools. Finchreach is a remote MCP server for LinkedIn at `https://mcp.finchreach.com/mcp`, authenticated with an API key header.

## Why draft LinkedIn posts with a local model?

Good posts usually start from raw material you wouldn’t paste into a cloud chatbot: meeting notes, a client retrospective, a messy brain dump about what went wrong this quarter. With a local model, those notes stay on your machine while the model turns them into a draft.

Be clear about what stays local and what doesn’t. The model and your notes run locally. Anything the model sends through a Finchreach tool, such as the text of a post draft or a search query, goes to Finchreach’s server and, once approved, to LinkedIn. Your notes only leave your machine if the model puts them into a tool call, which is one more reason to read every draft in the queue.

## Which MCP client should I use with Ollama?

| Client | How it uses Ollama | How it reaches Finchreach |
| --- | --- | --- |
| Goose | Ollama is a supported provider (official Ollama integration page) | Desktop: Extensions → Add custom extension → Streamable HTTP, with an `Authorization` request header |
| ollmcp (mcp-client-for-ollama) | A terminal client built for Ollama | A JSON config with `streamable_http`, URL and headers |
| Open WebUI | A web chat UI that commonly runs on Ollama | External Tools → MCP (Streamable HTTP) with a Bearer token |
| Claude Code or Codex CLI | Pointed at Ollama via the official integration pages | Their own remote MCP config with a header |

If you want a desktop app, start with Goose. If you live in the terminal, ollmcp is the lightest. Open WebUI suits a shared, self-hosted chat for a small team. Exact field names vary by client version, so check each client’s docs when you set it up, as of October 2026.

> Use a model that supports tool calling, such as qwen3 or llama3.1 and newer. Smaller models can call tools unreliably: they may skip the tool, call it with the wrong arguments or claim they posted when they didn’t. The approval queue catches bad drafts, but a stronger model saves you rejections.

## How do I set it up?

The full walkthrough is in the [Ollama guide](https://finchreach.com/guides/ollama), and the [Goose guide](https://finchreach.com/guides/goose) covers that client in detail. The short version with ollmcp:

1. [Create a Finchreach account](https://finchreach.com/sign-up), connect your LinkedIn account on the Accounts page, and create an API key (it starts with `lk_`).
2. Install Ollama and pull a tool-capable model, for example `ollama pull qwen3`.
3. Install your MCP client. For ollmcp, follow the [mcp-client-for-ollama README](https://github.com/jonigl/mcp-client-for-ollama).
4. Add Finchreach to the client’s config with the Streamable HTTP transport and your key in an `Authorization` header (example below).
5. Start the client with your model and ask it to list your Finchreach LinkedIn accounts to test the connection.

ollmcp server config:
```json
{
  "mcpServers": {
    "finchreach": {
      "type": "streamable_http",
      "url": "https://mcp.finchreach.com/mcp",
      "headers": { "Authorization": "Bearer <FINCHREACH_API_KEY>" }
    }
  }
}
```

For Goose with Ollama, follow [Ollama’s Goose integration page](https://docs.ollama.com/integrations/goose) to set the provider, then add Finchreach as a Streamable HTTP extension with the header. Our [quickstart](https://finchreach.com/docs) covers the Finchreach side for every client.

## What should I ask the local model?

Give the model your notes in the chat (or point a file-capable client at them), then ask for a draft:

> Ask your assistant: “Here are my notes from this week’s project retro. Turn the most interesting lesson into a LinkedIn post: a concrete opening line, the situation in two sentences, what we learned, and a question for readers. Under 1,000 characters. Leave out the client name and any numbers. Show me the draft here; don’t create anything yet.”

When the draft reads well:

> Ask your assistant: “Use Finchreach to read my LinkedIn profile so the voice matches my headline, adjust the draft if needed, then create the post with linkedin_create_post. It should go to my approval queue.”

To get ideas before you write:

> Ask your assistant: “Search LinkedIn posts about “fractional CFO” and summarize the three angles people keep repeating. Suggest one angle from my notes that nobody seems to cover.”

## Can a local model post to LinkedIn without me?

Not with approval mode on, which is the default. `linkedin_create_post` and every other write (comments, reactions, DMs, invitations) become pending actions. The client gets `pending_approval` and an action id. You read the exact text on the [approvals page](https://finchreach.com/approvals) and approve or reject it. Only approved actions run.

That matters more with local models, which are more likely to make tool-calling mistakes than large hosted ones. If a model tells you it posted, check the queue: the post is pending, not live.

> Automating a personal LinkedIn account carries risk under LinkedIn’s User Agreement. The approval queue lowers that risk because you review and pace every action, but it doesn’t remove it.

## Ollama and LinkedIn: common questions

Is there a native Ollama integration for Finchreach? No. Ollama has no MCP client, so you always need one of the clients above. That is how MCP works with Ollama in general, not a Finchreach limitation.

Does everything stay on my machine? The model and your notes do. Tool calls go to Finchreach, and approved actions go to LinkedIn. Finchreach also needs your LinkedIn connection to act for you.

Can the model post images? No. Finchreach supports text posts only, with no scheduling, company pages or analytics.

Want a hosted model instead? See [using OpenRouter models with Finchreach](https://finchreach.com/blog/openrouter-linkedin-notification-digest), or [Claude Code](https://finchreach.com/integrations/claude-code). New to MCP? Read [what a LinkedIn MCP server is](https://finchreach.com/blog/what-is-a-linkedin-mcp-server).

Draft privately, publish deliberately: [sign up for Finchreach](https://finchreach.com/sign-up) and follow the [Ollama guide](https://finchreach.com/guides/ollama).