Ollama MCP: run local models on live TikTok and Instagram data
Updated
Virlo is social listening for TikTok and Instagram Reels. A model running in Ollama only knows its training data, so it cannot say which TikTok sounds or Instagram Reels are rising this week. Pair that model with an MCP client such as goose, or with Ollama's own tool calling, and it can query Virlo for live trends, hooks and outlier videos while the model itself stays on your machine. Ollama is an open-source runtime for downloading and running large language models locally behind a simple API.
- goose on an Ollama model reaches Virlo over MCP with a browser sign-in
- Ollama's tool calling + the Virlo API for your own scripts
- Inference stays local; only the Virlo lookups leave your machine
Part of Virlo's AI assistant guides. Want a chat window on top of Ollama? See Open WebUI. Running the agent itself? See goose.
How do I use Ollama with an MCP server?
Ollama runs the model; an MCP client runs the conversation and calls Virlo. This route uses goose, an open-source agent that speaks MCP and lists Ollama as a provider.
Step 1: Create your Virlo account for free
Step 2: Pull a model that supports tool calling
Tool calls are what let a local model ask Virlo for data, so choose a model Ollama marks for tools. Ollama's own tool-calling examples use qwen3.
Terminalollama pull qwen3Step 3: Point goose at Ollama
Run the command below, choose Configure Providers, pick Ollama and type the model name. goose talks to Ollama on
http://localhost:11434unless you setOLLAMA_HOST.Terminalgoose configureStep 4: Add Virlo as a remote extension
Run
goose configureagain, choose Add Extension, then Remote Extension (Streamable HTTP). Name it virlo and paste the endpoint below. goose registers itself with Virlo over OAuth and opens the sign-in page in your browser the first time it connects.Endpointhttps://dev.virlo.ai/api/mcp/mcpStep 5: Optional: call Virlo from your own Ollama script
Skipping the agent? Export
VIRLO_API_KEYin your shell and hand Ollama a Python function that calls the Virlo API. The model returns a tool call, the script runs it and sends the JSON back for the final answer.virlo_ollama.py# virlo_ollama.py: Ollama tool calling + the Virlo REST API import os, requests, ollama def virlo_trends(platform: str = 'tiktok', limit: int = 5) -> str: """Get this week's trends to copy on TikTok or Instagram from Virlo.""" r = requests.get('https://api.virlo.ai/v1/content-trends', params={'platform': platform, 'limit': limit}, headers={'Authorization': 'Bearer ' + os.environ['VIRLO_API_KEY']}) return r.text messages = [{'role': 'user', 'content': 'What are the top TikTok trends to copy right now?'}] reply = ollama.chat(model='qwen3', messages=messages, tools=[virlo_trends]) for call in reply.message.tool_calls or []: messages += [reply.message, {'role': 'tool', 'tool_name': call.function.name, 'content': virlo_trends(**call.function.arguments)}] print(ollama.chat(model='qwen3', messages=messages).message.content)Step 6: Start a session and ask
Run
goose session. Your local Ollama model decides when to call Virlo, and goose shows each tool call as it happens.Your first promptUse Virlo to list the TikTok sounds breaking out this week in [your niche], then suggest which two would fit a [tone] video for [your brand] and why.
Just testing? goose session --with-streamable-http-extension "https://dev.virlo.ai/api/mcp/mcp" loads Virlo for a single session without saving it to your goose config.
goose vs Open WebUI vs a Python script: how should Ollama reach Virlo?
| Route | What runs on your machine | How Virlo is called | Best for |
|---|---|---|---|
| goose + Ollama | Ollama serving the model, goose as the agent | Remote Extension (Streamable HTTP) with OAuth | Agent-style research in the terminal or goose desktop |
| Open WebUI + Ollama | Ollama plus a self-hosted chat interface | External Tool Servers, type MCP (Streamable HTTP) | A ChatGPT-like window for a whole team |
| Ollama tool calling | Ollama and your own Python or JavaScript code | A function you define that hits api.virlo.ai/v1 | Scheduled jobs and custom pipelines |
| LibreChat or LM Studio | Another local chat client | Their own MCP settings | If your team already uses one of them |
Ollama itself is the model runner in every row; the MCP connection always lives in the client that sits on top of it.
Which TikTok sounds would your Ollama model see today?
When a local model calls Virlo for trending TikTok sounds, these are the rows that come back, ranked by how many videos used each sound recently.
In Virlo's data, 316 TikTok videos used "original sound - hazel gray" in the last 7 days, more than any other sound on TikTok as of November 19, 2026.
As of November 19, 2026 at 12:03 UTC
| Sound | Artist | TikTok videos (7 days) | Licensed for brands |
|---|---|---|---|
original sound - hazel gray | hazel gray | 316 | Yes |
Flower In A Hurricane (feat. Tame Impala, SZA, King Gizzard & the Lizard Wizard) [From GTAVI: Original Score] | Justice & Tame Impala & SZA & King Gizzard & the Lizard Wizard | 287 | Yes |
original sound | spam_lilah1 | 276 | Yes |
| original sound - cluberich | cluberich | 264 | Yes |
| original sound - rosyconfessions | ᝰ.ᐟ | 252 | Yes |
See all 300 trending TikTok sounds
Usage history and the TikTok videos behind every sound are in the Virlo app.
Open sounds in VirloHow this sample is made
- Source
- Virlo's sound rankings for TikTok, built from the TikTok videos Virlo collects.
- Window
- TikTok videos posted in the last 7 days that Virlo collected. Counts are TikTok videos in Virlo's data, not every video on the platform.
- Method
- Ranked by the number of videos using each sound in the window. "Licensed for brands" = the sound is in TikTok's commercial music library.
- As of
- November 19, 2026 at 12:03 UTC. Rebuilt every day.
- Made by
- Numbers are pulled automatically from Virlo's data; the text around them is written by the Virlo team.
Suggested citation: Source: Virlo, “Ollama MCP: run local models on live TikTok and Instagram data”, virlo.ai/ollama
How does Ollama tool calling work with the Virlo API?
Ollama's /api/chat endpoint accepts a list of tools. When the model wants data, it returns a tool_calls entry instead of text; your code runs the matching function, appends a tool message with the result and asks again. The Python library builds each tool's schema straight from a function's type hints and docstring.
That makes Virlo easy to wrap. Export VIRLO_API_KEY in your shell, define one function per Virlo endpoint you need, and pass them as tools=[...]. The optional script in the setup above gives a local model this week's trends to copy; add functions for sounds, hashtags or creators the same way.
| Piece | What it does |
|---|---|
ollama pull qwen3 | Downloads a model with tool calling |
def virlo_trends(...) | Calls GET https://api.virlo.ai/v1/content-trends with your bearer key |
tools=[virlo_trends] | Lets the model decide when to call it |
tool_calls loop | Runs the function and feeds the JSON back as a tool message |
Install the two libraries with pip install ollama requests. The JavaScript library follows the same pattern, and any language with an HTTP client can call Virlo.
Why marketers run Virlo research on Ollama
Teams pick Ollama when prompts, client names or drafts should not leave their hardware. With Virlo attached, only the lookup itself goes out: a keyword, a handle or a video link. The analysis, the scripts and the notes are written by the model on your own machine.
It also suits long, repetitive jobs. A local model can loop over twenty competitors or a hundred hooks without anyone watching a chat window, using Virlo as the source of facts each time.
What if the Ollama model never calls Virlo?
The connection can be perfect and the model can still answer from memory. Local models vary widely in how well they follow tool instructions, so try these in order:
- Start the prompt with 'Use Virlo to...' so the request clearly needs a tool.
- Check the model supports tools in the Ollama library before blaming the setup.
- Move to a larger variant of the same model family, which usually follows tool schemas more reliably.
- Raise the context length so long Virlo results are not cut off mid-answer.
- In goose, confirm the virlo extension is enabled for the session you are using.
Can Open WebUI or LibreChat replace goose for Ollama users?
Yes, if your team prefers a browser chat to a terminal agent. Both connect to an Ollama server as a model provider and both add Virlo as a Streamable HTTP tool server, so the Ollama side of the setup does not change. The Open WebUI and LibreChat guides cover each one.
What can a local Ollama model ask Virlo?
Use these in goose, Open WebUI or your own script with an Ollama model. Replace the brackets.
Sound shortlist
Use Virlo to list breakout sounds on TikTok and Instagram Reels this week and pick three that suit [your brand].
Private brief
Pull the top 5 TikTok trends to copy in [your niche] from Virlo and write an internal brief for [client] that stays on this machine.
Competitor sweep
For each of @[competitor 1], @[competitor 2] and @[competitor 3], use Virlo to find their videos that beat their usual views and summarise the pattern.
Hook rewrite
Get 15 proven hooks in [your niche] from Virlo and rewrite each for [your offer] in our house style.
Transcript study
Fetch the transcript of [TikTok link] through Virlo and break it into hook, setup, payoff and call to action.
Creator list
Find 10 creators under [N] followers in [your niche] whose videos outperform their follower count, using Virlo, and save the list as markdown.
Hashtag check
Which hashtags are growing on Instagram around [topic] according to Virlo, and which ones look saturated?
Ollama vs LM Studio vs Open WebUI: which local tool is which?
Local AI tools stack on top of each other, so the names get mixed up.
Ollama (this page)
The model runner: it downloads models and serves them on localhost. It needs a client such as goose or a script to call Virlo.
LM Studio
A desktop app that runs models and has its own MCP support in mcp.json, so it reaches Virlo without a separate client.
Open WebUI
A self-hosted chat interface that often sits on top of Ollama and connects to MCP servers from its admin settings.
goose
An open-source agent that can use an Ollama model and add Virlo as a Streamable HTTP extension.
What are the limits?
- Virlo answers with public TikTok and Instagram data and cannot post; a local model can research and draft with it, nothing more.
- Small local models sometimes skip tools or send odd arguments. If the model ignores Virlo, move up to a larger tool-calling model.
- The Virlo connection lives in the client (goose, Open WebUI, LibreChat) or in your script, never inside Ollama itself.
- Trends to copy refresh hourly and are checked by the Virlo team; a local job that runs more often than that will see the same list.
- Plan details are on pricing.
Frequently asked questions
Does Ollama support MCP?
Ollama itself runs models and supports tool calling, while MCP connections are handled by the client in front of it. Clients such as goose and Open WebUI can use an Ollama model and connect to remote MCP servers like Virlo, so the pair together gives a local model MCP tools.
How do I connect an MCP server to Ollama?
Use an MCP client that supports Ollama as its provider. In goose, run goose configure, set Ollama as the provider, then Add Extension, Remote Extension (Streamable HTTP), and paste https://dev.virlo.ai/api/mcp/mcp. goose signs you in to Virlo and the Ollama model can call its tools.
What is the best Ollama MCP client?
It depends on how you work. goose suits agent-style research in a terminal or desktop app, Open WebUI gives a team a shared chat interface, and LibreChat fits companies that manage configuration in YAML. All three can run an Ollama model and reach Virlo over Streamable HTTP.
Which Ollama models work with tools?
Models that support tool calling, which Ollama tags in its library. Ollama's own tool-calling examples use qwen3. Larger models follow tool instructions more reliably, so if a small one keeps answering without calling Virlo, try a bigger variant. Raising the context window also helps fit longer Virlo results.
Can Ollama call an API?
Not on its own, but its tool calling lets your code do it. Define a function that calls the Virlo API, pass it to ollama.chat as a tool, run it when the model returns tool_calls, and send the result back as a tool message. The model then answers using that data.
Can a local LLM see TikTok trends?
Yes, once it can reach Virlo. Through an MCP client or a tool function, a local model can read TikTok trends to copy, outlier videos ranked by Virality Score, sounds, hashtags, creators and transcripts, then reason about them on your hardware.
Does my data leave my machine when Ollama uses Virlo?
Only the lookups do. When the model calls Virlo it sends the query, such as a keyword or creator handle, and receives public TikTok and Instagram data back. Prompts, drafts and the model itself stay local in Ollama. Run the client on the same machine to keep it that way.
Can Ollama post to TikTok through Virlo?
No. Virlo is read-only research and never connects to social accounts, so an Ollama setup can study and write with it while posting stays in your normal tools. That also means no social passwords are ever stored on the machine running Ollama, so the setup stays simple to secure, even on a shared office computer.
Ollama MCP server for TikTok: is there one?
Yes, in effect. Virlo's hosted MCP server gives any Ollama-backed client read-only tools for TikTok and Instagram research: trends to copy, outlier videos, hooks, sounds, hashtags, creators and transcripts. Add it in goose, Open WebUI or LibreChat, choose your Ollama model, and the model calls those tools as needed.
Give your local model a live view of social media
Keep Ollama on your machine and let it ask Virlo what is working this week.
More AI assistant guides
- AI assistant guides
- Claude
- Claude Code
- Cursor
- MCP server
- TikTok MCP server
- Instagram MCP server
- ChatGPT
- ChatGPT dots
- OpenAI Codex
- Gemini CLI
- Grok Build
- Claude Cowork
- n8n
- Zapier
- Grok Bot
- Muse
- Claude Duet
- Hermes Agent
- Kimi K3
- Cue
- Manus
- Perplexity Computer
- Paperclip
- OpenClaw
- NanoClaw
- DeepSeek
- AI agents CLI
- ChatGPT agent
- Perplexity
- Microsoft Copilot
- Le Chat
- Notion
- Slack
- Discord
- Telegram
- Google Sheets
- Windsurf
- VS Code
- GitHub Copilot
- Cline
- JetBrains
- Kiro
- Google Antigravity
- goose
- Warp
- OpenCode
- Zed
- Amp
- Factory Droid
- Raycast
- Lovable
- Replit
- Bolt.new
- v0
- LM Studio
- Open WebUI
- LibreChat
- Make
- Pipedream
- Gumloop
- Lindy
- Relevance AI
- Airtable
- ClickUp
- Dify
- Flowise
- LangChain
- CrewAI
- OpenAI Agents SDK
- Mastra
- Vercel AI SDK
- Claude Agent SDK

![Cover art for Flower In A Hurricane (feat. Tame Impala, SZA, King Gizzard & the Lizard Wizard) [From GTAVI: Original Score]](https://media.virlo.ai/sound-covers/f59fe6c5233489f9b8bc083eb73ea41509e58c05fd5f7fa96ce5e016727bc458.jpg)
