# Forefront MCP for AI Agents AI Agent Connect > Forefront lets you talk to your models, run fine-tuning jobs, and manage data pipelines directly from your AI client. It's built for anyone who needs to move beyond simple prompting into actual model customization and large-scale output collection. ## Overview - **Category:** developer-tools - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_9p3sy2z5j8vwdMtOomVYymmFfYsP4YghbaGI6v8C/ai-agent-connect - **Tags:** llm, fine-tuning, text-generation, pipelines, ai-models ## Description This Connector connects your Forefront account to your AI agent so you can manage models without leaving your workspace. It handles the heavy lifting of fine-tuning, data collection, and pipeline management through a simple chat. Instead of jumping between tabs to check training statuses or manually logging outputs, you can just ask your agent to do it. You can kick off a training job, check your pipeline counts, or organize your datasets with a few words. It turns your AI client into a command center for your model lifecycle. You'll find this especially useful when you're using Vinkius to manage your entire AI stack, as it keeps your workflow focused on the models rather than the manual overhead of the API. You can move from an idea to a trained model much faster because the agent handles the repetitive tasks of data organization and job monitoring. It removes the friction of manual data entry and lets you focus on the actual quality of your model outputs. Whether you're building a custom chatbot or refining a specialized tool, this Connector gives you a direct line to your training data and model performance metrics. You can automate the collection of thousands of samples, group them into specific datasets, and monitor the health of your fine-tuning runs in real time. It replaces the need for complex scripts and manual spreadsheet tracking with a conversational interface that actually understands your data needs. ## Tools ### create_pipeline_dataset Turn a selection of pipeline data into a new dataset. This is great for prepping training sets. ### create_fine_tune Kick off a new fine-tuning job. You can use your own training and validation data. ### create_pipeline Create a new pipeline to collect LLM outputs. It's the best way to gather samples at scale. ### get_pipeline_count Get the count of items in a pipeline selection. Use this to check your progress quickly. ### get_pipeline_samples Get the actual data samples from a pipeline. This lets you inspect your model's performance. ### get_pipeline Get the full details of a specific pipeline. Use this to see the metadata for any ID. ### add_pipeline_data Add new data samples to a specific pipeline. This keeps your training logs organized. ### list_pipelines See a full list of your active pipelines. It helps you stay on top of multiple runs. ### create_chat_completion Generate a response using the chat-ml format. This is the standard for multi-turn conversations. ### create_completion Generate a response from a single prompt. Use this for simple, one-off text tasks. ## Prompt Examples **Prompt:** ``` I need to start a new fine-tuning job for my customer service model. ``` **Response:** ``` I'll get that started for you. I'll use `create_fine_tune` to kick off the job using your current training dataset. **Job Details:** - **Model:** Forefront-Base - **Dataset:** `cs_v2_training_set` - **Status:** Initializing... I'll let you know as soon as the training job is active. ``` **Prompt:** ``` Can you show me all my active pipelines and how many samples are in each? ``` **Response:** ``` Here are your active pipelines: | Pipeline Name | ID | Sample Count | | :--- | :--- | :--- | | `marketing_copy_test` | `pipe_789` | 450 | | `customer_feedback_loop` | `pipe_123` | 1,204 | | `qa_testing_pipeline` | `pipe_456` | 89 | Would you like to see the samples for any of these? ``` **Prompt:** ``` Create a pipeline called 'marketing_copy_test' so I can collect some results. ``` **Response:** ``` Done! I've created a new pipeline for you. - **Name:** `marketing_copy_test` - **ID:** `pipe_789` Your agent is now ready to collect and organize any marketing copy outputs you generate. ``` ## Capabilities ### Create chat completions Generate responses using the chat-ml format for multi-turn conversations. ### Start fine-tuning jobs Kick off custom training jobs on base models using your own datasets. ### Build data pipelines Create automated pipelines to collect and organize LLM outputs at scale. ### Track pipeline samples Retrieve and inspect specific data samples from your active pipelines. ### List active pipelines Get a high-level view of all your active pipelines and their current status. ### Add data to pipelines Log new data samples directly into a specific pipeline for easy collection. ### Create datasets from pipelines Convert your collected pipeline data into structured datasets for training. ## Use Cases ### Automated sample collection A developer needs to create 1,000 chat samples for a new bot. They use create_pipeline to capture every response automatically. ### Training data auditing An ML engineer wants to see if their training data is balanced. They use get_pipeline_samples to inspect the logs. ### Remote job scheduling A data scientist needs to kick off a training run on a weekend. They use create_fine_tune to start the job via a quick prompt. ### Experiment overview A researcher wants to see the status of multiple experiments. They use list_pipelines to get a bird's eye view of all active runs. ## Benefits - Start fine-tuning jobs instantly with create_fine_tune without leaving your chat. - Organize your model outputs automatically using create_pipeline. - Keep your training data clean by using add_pipeline_data to log samples. - Quickly audit your progress with list_pipelines and get_pipeline_count. - Convert raw model outputs into training sets using create_pipeline_dataset. - Switch between chat-ml and standard prompts easily with create_chat_completion and create_completion. ## How It Works The bottom line is you get a direct interface for managing your full model lifecycle from a single chat window. 1. Subscribe to the Forefront MCP on Vinkius. 2. Enter your Forefront API Key in the connection settings. 3. Ask your agent to generate text, start training, or manage data pipelines. ## Frequently Asked Questions **How does the Forefront MCP help with training?** It lets you start and manage fine-tuning jobs directly from your AI client. You can kick off training runs and monitor their progress without switching to a separate web dashboard. **Can I use the Forefront MCP to manage my data pipelines?** Yes, you can create, list, and track data pipelines. This makes it much easier to organize large amounts of model outputs for later use. **Does the Forefront MCP support custom fine-tuning?** It does. You can use it to start fine-tuning jobs on base models using your own specific training and validation datasets. **How do I collect model outputs using the Forefront MCP?** You can create a pipeline to automatically collect and organize LLM outputs. This saves you from having to manually copy and paste results into a spreadsheet. **Can I see my training samples through the Forefront MCP?** Yes, you can retrieve specific data samples from any of your active pipelines. This is great for auditing your model's performance during development. **Is the Forefront MCP good for large scale data collection?** It's ideal for that. By using pipelines, you can gather thousands of samples at scale and then convert those selections into structured datasets for training. **How can I generate a model response using a chat conversation?** You can use the `create_chat_completion` tool. Provide the model name and an array of messages in chat-ml format to receive the generated response. **Can I start a custom fine-tuning job directly from my agent?** Yes! Use the `create_fine_tune` tool by specifying the name of your fine-tuned model, the baseModel, and the trainingDataset ID to begin training. **How do I collect LLM outputs using pipelines?** First, create a pipeline using `create_pipeline`. Once created, you can use `add_pipeline_data` to log messages, user IDs, and custom metadata directly into that pipeline.