How do I connect Helicone to my AI client?

You subscribe to the MCP on Vinkius and provide your Helicone API Key. Once connected, you can use it immediately in Claude, Cursor, or any MCP-compatible client.

Can I track costs per user?

Yes. You can ask your agent to break down spending by user, model, or custom metadata properties.

What latency metrics can I see?

You can measure Time To First Token (TTFT) to identify which upstream providers are causing delays.

Does this work with my existing prompts?

Yes. You can access managed prompt versions and track changes to your instruction logic through the agent.

How do I use user feedback to improve my models?

The MCP allows you to extract user critiques like thumbs up or down ratings to help improve model grounding.

Can I see the exact prompt that caused a specific error?

Yes. Use the query_requests capability to fetch direct prompts and outputs from the proxy logs. You can filter by status or custom tags to find the exact interaction that needs debugging.

How do I track costs for a specific customer ID?

Ask your agent to query_costs and include your customer identity in the filter. Helicone maps costs per model and user, allowing you to see exactly how much each client is burning in LLM tokens.

Can my agent log human feedback into Helicone?

Absolutely. Use the log_feedback capability to inject offline Human-in-the-Loop verdicts or text critiques directly into Helicone's database, helping you refine your model's grounding over time.

An astronaut overseeing a futuristic data hub

Download PDF Markdown Version

Connect Helicone LLM Observability to ChatGPT and Claude

Set it up once and it works in ChatGPT, Claude, Cursor, and any other client you love. It all runs safely in the background, so you can relax and focus on what really matters. Watch your AI costs and performance in real time.

Ask AI about this Connector

Developed, maintained, and hosted by Vinkius.

MCP VERIFIED · PRODUCTION READY · VINKIUS GUARANTEED

Waiting for input…

Works with modern AI clients that support MCP, including ChatGPT, Claude, Cursor, and more.

Your AI does the work. We keep it safe and sound.

No other AI tool gives you this. Your workflow runs in the client you already love, your account stays protected, and you decide what the agent can and cannot do. It is the whole reason to make the switch.

Each Connector lives inside our infrastructure, the same place where your workflow runs. The three things below are already part of it, working quietly in the background so you never have to think about them. Nothing to add, nothing to set up, nothing to worry about.

V8 Isolate01

Each Connector runs inside its own sealed environment.

Each Connector runs inside its own sealed V8 runtime — no access to your files, your system, or other servers. A Connector never sees past its own request.

Kill Switch02

One switch stops every agent. Instantly.

One action deactivates every Connector, revokes every token, and ends every session at once. AI Governance keeps the record.

FinOps per User03

Spending limits that pause instead of surprise.

Every user keeps their own connections and credentials, isolated from everyone else. Anything that would cross your spending limit is held for your approval, not billed.

Three controls, built into the layer your AI connects through.

Curated set · 10 capabilities

We ship only the actions that matter.

Capability set01 / 03

01-04

4 capabilities in this set.

Part of 10 available through Helicone.

01
Query sessions
This capability pulls active billing data and structured rules from your sessions.
02
Log feedback
Use this to identify active arrays and handle native parsing for feedback logs.
03
Query feedback
This capability inspects internal arrays to help you mitigate specific plan math issues.
04
Query latency
Get a JSON payload containing latency data and customer bindings.

Capability set02 / 03

05-07

3 capabilities in this set.

Part of 10 available through Helicone.

05
Query requests
This capability identifies bounded CRM records within the Helicone platform.
06
Query prompts
Access cloud logging and tracing to check explicit vault limits for prompts.
07
List properties
This capability identifies active arrays related to native gateway authentication.

Capability set03 / 03

08-10

3 capabilities in this set.

Part of 10 available through Helicone.

08
Get prompt versions
Retrieve specific prompt versions and extract churn flags.
09
Query costs
This capability performs structural extraction of properties that drive account logic.
10
Query users
Use this to run automated validation checks on gateway history.

Real work

What teams actually run

Budget Management

Check how much a specific feature is costing you in real time.

Performance Tuning

Identify which LLM provider is causing high TTFT for your users.

Prompt Iteration

Compare different versions of a prompt to see how changes affect output.

Error Debugging

Inspect exact prompts and outputs from proxy logs to find where logic failed.

Automate a task

Put Helicone to work on one thing.

Step-by-step automations your AI runs through this Connector: the workflow, the prompts to start with, and the exact capabilities behind each move.

One connector, every AI

Helicone works with the most popular AI clients.

These are the most popular clients, each with a step-by-step guide: one link, set up once, with governance and visibility built in. And because everything runs on the MCP standard, the same connection also works in any other compatible client — nothing to rebuild.

Building your own app? The connector is yours to use.

You don't need a client to put Helicone to work: the same hosted connection plugs into your own applications and agent code, with the same governance on every request. Build with it, chat with it — one connection for both.

Observed, not estimated

930ms average. Fast in production.

Helicone is checked daily against the live service.

Daily averagePeak 1197ms

Oct 3Today

Fastest day: 745ms
Slowest day: 1197ms
14-day trend: Slowing+15%

Connect your client

One URL. Every client.

Activate the Connector, copy your link, and paste it into the client you already use. 10 capabilities arrive ready to run.

Preview access · not provider authentication

The vk_preview_* token belongs to Vinkius preview infrastructure. It lets Claude discover and display the capabilities of Helicone, so you can see the experience inside your AI.

It does not authenticate your account with Helicone. Actions requiring credentials or live account data may not run until you activate the Connector and authorize the service.

Helicone Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_BY11i3Opi4Qq9KtIJUGQpInAaXPfnwFfbeNDvYuq/mcp

Claude Desktop

Follow the steps below to connect in seconds.

1In Claude Desktop, open Settings → Connectors.
2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
3Click Add and start a new chat — Helicone capabilities are ready to use.

Configuration · claude_desktop_config.jsonCopy

{
  "mcpServers": {
    "helicone-llm-observability-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_BY11i3Opi4Qq9KtIJUGQpInAaXPfnwFfbeNDvYuq/mcp"
    }
  }
}

Claude
ChatGPT
Cursor
VS Code
Windsurf
Claude Code
JetBrains
Cline

Step-by-step instructions for each client are in the guide. How to connect

Guided setup for Claude? Set up Helicone in Claude and other AIs

See all the AI clients this connector works with ↑

Who it's for

Built for the work Helicone owners hand off.

This MCP is for technical teams running LLM-based applications in production.

01
LLM Engineers
They use it to debug prompt performance and measure TTFT across providers.
02
Product Owners
They monitor AI spending and calculate costs per user or feature.
03
Data Scientists
They analyze user feedback and logged critiques to improve response quality.
04
DevOps/SREs
They ensure the reliability of the AI gateway and proxy layers.

FAQ

Questions Helicone owners ask.

01
How do I connect Helicone to my AI client?
You subscribe to the MCP on Vinkius and provide your Helicone API Key. Once connected, you can use it immediately in Claude, Cursor, or any MCP-compatible client.
02
Can I track costs per user?
Yes. You can ask your agent to break down spending by user, model, or custom metadata properties.
03
What latency metrics can I see?
You can measure Time To First Token (TTFT) to identify which upstream providers are causing delays.
04
Does this work with my existing prompts?
Yes. You can access managed prompt versions and track changes to your instruction logic through the agent.
05
How do I use user feedback to improve my models?
The MCP allows you to extract user critiques like thumbs up or down ratings to help improve model grounding.
06
Can I see the exact prompt that caused a specific error?
Yes. Use the query_requests capability to fetch direct prompts and outputs from the proxy logs. You can filter by status or custom tags to find the exact interaction that needs debugging.
07
How do I track costs for a specific customer ID?
Ask your agent to query_costs and include your customer identity in the filter. Helicone maps costs per model and user, allowing you to see exactly how much each client is burning in LLM tokens.
08
Can my agent log human feedback into Helicone?
Absolutely. Use the log_feedback capability to inject offline Human-in-the-Loop verdicts or text critiques directly into Helicone's database, helping you refine your model's grounding over time.

Explore

More in AI Frontier

Connect Helicone to Claude, Cursor, ChatGPT & more

Your AI does the work. We keep it safe and sound.

Each Connector runs inside its own sealed environment.

One switch stops every agent. Instantly.

Spending limits that pause instead of surprise.

We ship only the actions that matter.

01-04

Query sessions

Log feedback

Query feedback

Query latency

05-07

Query requests

Query prompts

List properties

08-10

Get prompt versions

Query costs

Query users

What teams actually run

Budget Management

Performance Tuning

Prompt Iteration

Error Debugging

Building your own app? The connector is yours to use.

Helicone Connector

Claude Desktop

LLM Engineers

Product Owners

Data Scientists

DevOps/SREs

How do I connect Helicone to my AI client?

Can I track costs per user?

What latency metrics can I see?

Does this work with my existing prompts?

How do I use user feedback to improve my models?

Can I see the exact prompt that caused a specific error?

How do I track costs for a specific customer ID?

Can my agent log human feedback into Helicone?

LangSmith (LLM Observability & Hub) AI Connector

New Relic AI (LLM Observability) AI Connector

Humanloop (LLM Prompt Management API) AI Connector

Portkey AI Connector

Suggestions

Loggly (Cloud Log Management API) AI Connector

Langfuse (LLM Tracing & Evals) AI Connector

Fuel Economy Log Analysis AI Connector

LiteLLM (LLM Proxy & Spend Tracking) AI Connector