# Helicone MCP connector

The Helicone connector brings your Helicone requests, prompts, evals, and usage dashboards to Claude, ChatGPT, Cursor, and the Elaichi Agent, so anyone on the team can ask about what their AI application is doing and act on it from the tool they already use.

Source: https://elaichi.ai/connectors/helicone/

## Facts

| | |
| --- | --- |
| Application | Helicone |
| Category | Analytics |
| AI tools | 46 |
| Authentication | Connects with an API key |
| Needs your own OAuth app | No |
| MCP endpoint | https://api.elaichi.ai/mcp |
| Works with | Claude, ChatGPT, Cursor, any MCP client, and the Elaichi Agent |
| Tools advertised by name | No. 46 tools is past the 30-tool threshold, so clients use `search_tools` and `execute_tool` |

## What you can ask once Helicone is connected

- Summarize Helicone requests with low feedback scores this week.
- Which team created the most Helicone requests yesterday?
- Show scores on Helicone requests tagged production.

## Connect Helicone in Elaichi

This happens once for the organization, before any client is involved.

1. Open Connections, choose Add connection, and pick Helicone.
2. Optionally set Share with, then press Connect.
3. Paste a Helicone API key. One person generates a token in Helicone and pastes it once. Everyone else works through Share with, and never sees it.

Credentials are vaulted and nobody, including the AI, reads them back. The connection becomes a toolbox immediately, so you can curate which Helicone tools are exposed, rename them, or freeze arguments before anyone points a client at it.

## Connect Helicone to Claude

Endpoint: https://api.elaichi.ai/mcp

1. Open Customize, then Connectors.
2. Press Add.
3. Name it, paste the MCP server URL, then Continue.
4. Sign in and approve.

On Team and Enterprise, an Owner adds it once. Everyone else turns it on for themselves.

## Connect Helicone to ChatGPT

Endpoint: https://api.elaichi.ai/mcp

1. Open Plugins, then press the + button.
2. Name it and paste the endpoint into Server URL.
3. Leave Authentication on OAuth, then tick the risk acknowledgement.
4. Press Create, then sign in and approve.

Works on the web today. The plugin directory lives at chatgpt.com/plugins.

## Connect Helicone to Cursor

Endpoint: https://api.elaichi.ai/mcp

1. Open `~/.cursor/mcp.json`.
2. Add the endpoint under `mcpServers`.
3. Reload Cursor, then sign in and approve.

Set up per machine, so repeat it on each computer you work from.

## Connect Helicone to any MCP client

Endpoint: https://api.elaichi.ai/mcp

1. Add the endpoint as a remote MCP server.
2. Sign in and approve.

The Elaichi Agent already has these tools, with nothing to set up.

## What the consent screen decides

Only Read is granted by default, which is not enough to call a Helicone tool. Over MCP there is no trusted place to confirm a write in the moment, so the consent screen is the standing approval rather than a formality. Grant Read and Run tools. Think hard before granting Delete, which reaches into connected apps and cannot be undone.

## What teams do with Helicone through Elaichi

### Find the request behind a bad answer

Engineering. When a user reports a strange response, pull up the exact Helicone request by ID, see the inputs that went in, and tag it with a property so the whole team can find it again.

### Score responses without leaving the chat

Product. Review a batch of recent Helicone requests, submit a score or a note of feedback on each one, and build up the signal your evals depend on.

### See how eval scores are distributed

Data. Ask how a Helicone eval is scoring across the last week, spot the requests that fall at the bottom, and compare against the models in use.

### Check AI spend before the invoice lands

Finance. Read the Helicone dashboard figures for requests and cost by model or by property, and summarize them for the monthly budget review.

### Trace a customer's session end to end

Support. List the Helicone requests tied to a customer or session property, read what was asked and answered, and attach feedback so engineering knows what to fix.

### Keep prompts tidy across versions

Engineering. Create a new Helicone prompt, look up an existing one, and remove old versions that nobody should be calling any more, all from a short ask.

## Frequently asked questions

### How do I connect Helicone to Claude?

First, connect Helicone in Elaichi by pasting in a Helicone API key, which you can copy from your Helicone settings. There is no OAuth application to register and no client ID or secret to generate. Then in Claude open Customize, then Connectors, then Add, and paste the endpoint https://api.elaichi.ai/mcp. Sign in as yourself and Helicone requests, prompts, and evals are available in your conversation.

### Does Helicone work with ChatGPT and Cursor as well as Claude?

Yes. Once Helicone is connected in Elaichi, the same endpoint, https://api.elaichi.ai/mcp, works in Claude, ChatGPT, Cursor, any other MCP client, and the Elaichi Agent. You connect Helicone once and every client picks it up, with the same access rules and the same audit log.

### What can an AI agent actually do with my Helicone data?

It can look up a specific Helicone request and show you the inputs and the response, list recent requests and filter them by a property such as a customer or session, and read the dashboard numbers for usage and cost. It can also submit feedback or a score on a request, create an eval and show how its scores are distributed, list the models and properties in use, and create or retire prompts. Because Helicone exposes a large set of actions, short concrete asks such as "show request abc123" work better than long paragraphs.

### Does connecting Helicone give the AI everything in my organization?

No. Access follows the person who signed in, so the AI can only see the Helicone requests, prompts, and evals that the API key behind the connection can already see. Elaichi can narrow that further, for example read only for a Support team, but it can never widen it beyond what Helicone itself allows.

### Can my team share one Helicone connection?

Yes. One person connects Helicone with an API key and shares the connection with a team in Elaichi, and nobody else on the team ever handles or sees that key. Each teammate still signs in to Elaichi as themselves, so the audit log names the actual person who looked up a Helicone request or scored a response.

### Can I stop an agent from deleting or changing things in Helicone?

Yes. Restrictions in Elaichi apply per action, so you can allow reading Helicone requests and dashboards while blocking deleting prompts, submitting scores, or changing properties. A blocked action is never advertised to Claude, ChatGPT, or Cursor at all, so no prompt, however it is phrased, can reach it.

### What happens to a Helicone connection when someone leaves?

Offboarding that person in Elaichi ends their access to Helicone at once, across every client they used. If they had shared a Helicone connection with a team, it keeps working for everyone else. If you want Helicone gone entirely, disconnecting it once in Elaichi removes it from Claude, ChatGPT, Cursor, and every other client in the same step.

## All 46 Helicone tools

Every tool below is callable through https://api.elaichi.ai/mcp once Helicone is connected, subject to the toolbox it is in and the restrictions on the caller.

- **Helicone AI gateway create chat completion** (Create). Create a chat completion via the Helicone AI Gateway, routing the request to the specified model with pass-through billing. Returns the OpenAI-compatible completion response from the routed provider; response shape is provider-specific. Required: model, messages. In BYOK mode the standard OpenAI Chat Completions schema is allowed.
- **Helicone AI gateway create response** (Create). Create a response via the Helicone AI Gateway using the OpenAI Responses API format, routing to the specified model with pass-through billing. Returns the OpenAI-compatible response from the routed provider; response shape is provider-specific. Required: model, input. In BYOK mode the standard OpenAI Responses API schema is allowed.
- **List all Helicone AI gateway** (List). List all AI models available through the Helicone AI Gateway. Returns: object, data, error. OpenAI-compatible /v1/models endpoint for discovering which models are available for routing.
- **Helicone AI gateway list multimodal** (List). List all multimodal AI models supported by the Helicone AI Gateway. Returns: object, data, error. This is an OpenAI-compatible endpoint for discovering models that support multiple input or output modalities (e.g., text + images).
- **Get single Helicone request by ID** (Get). Retrieve a single LLM request record by its unique ID, including the prompt, response, model, tokens used, latency, cost, and metadata.
- **Helicone requests get inputs** (Get). Fetch the input variables and template parameters associated with a specific request.
- **Helicone requests submit feedback** (Action). Submit a thumbs-up or thumbs-down feedback rating for a specific request to track user satisfaction.
- **Helicone requests submit score** (Action). Attach a named numeric evaluation score (e.g. from a human reviewer or automated evaluator) to a specific request.
- **Helicone requests submit assets** (Action). Upload or attach a file asset (e.g. an image or document) to a specific request by asset ID.
- **Helicone requests upsert property** (Upsert). Add or update a custom key-value property on a request, useful for tagging requests with application-specific metadata.
- **List all Helicone requests** (List). Query and list LLM requests with filters for model, time range, user, properties, and scores. Returns paginated results with full request details.
- **Helicone requests bulk list** (List). Query large volumes of requests using ClickHouse for high-performance analytics. Supports the same filters as list but optimized for bulk data exports.
- **Helicone requests bulk get** (Get). Fetch multiple request records in a single call by providing an array of request IDs.
- **List all Helicone dashboard** (List). List dashboard scoring metrics in Helicone filtered by time range, user filters, and scoring criteria. Returns score metric records from the data array with configuration-specific attributes. Supports filtering by model, provider, cost, latency, tokens, custom properties, scores, and many other request/response attributes.
- **Create a Helicone eval** (Create). Create an evaluation score for a specific request in Helicone. Returns: data, error. Required: eval_id, score, name.
- **Helicone evals query score distributions** (Search). Query eval score distributions in Helicone using flexible filters on request/response metadata and time range. Returns: data, error.
- **List all Helicone evals** (List). List evaluation scores from Helicone. Returns: a list of evaluation score string values. No required parameters.
- **Helicone evals search** (Search). Search Helicone evaluation scores with request/response filters. Returns: name, averageScore, minScore, maxScore, count, overTime, averageOverTime. All filter and timeFilter fields are optional.
- **List all Helicone models** (List). List all AI models from the Helicone model registry with their configurations, pricing, and provider capabilities. Returns: id, error.
- **List all Helicone properties** (List). List properties in Helicone with pagination. Returns: property (a property name string for each record). No required parameters.
- **Create a Helicone prompt** (Create). Create a new prompt in Helicone with a name, tags, and an initial prompt body following the OpenAI chat completion format. Returns: id, versionId. Required: name, tags, promptBody.
- **Get single Helicone prompt by ID** (Get). Get a single Helicone prompt by id, retrieving its metadata including name and tags. Returns: id, name, tags, created_at. Required: id.
- **Delete a Helicone prompt by ID** (Delete). Delete a Helicone prompt by id, permanently removing the prompt and all associated versions. Returns an empty 204 response on success. Required: id.
- **Helicone prompts delete version** (Delete). Delete a specific version of a Helicone prompt while keeping the prompt and other versions intact. Returns an empty 204 response on success. Required: prompt_id, version_id.
- **Helicone prompts get count** (Get). Get the total count of prompts in the Helicone organization. Returns the count as an integer.
- **Helicone prompts get environments** (Get). List all environment names used across prompt versions in Helicone. Returns an array of environment name strings such as production, staging, and development.
- **Helicone prompts get tags** (Get). List all unique tags used across prompts in Helicone. Returns an array of tag strings such as support, chatbot, and classification.
- **Helicone prompts get version inputs** (Get). Get the input variables used when a specific prompt version was executed in a Helicone request. Returns: request_id, version_id, inputs. Required: prompt_id, version_id, request_id.
- **Helicone prompts get body** (Get). Get the complete prompt body content for a Helicone prompt version, including messages, model configuration, and parameters. Returns: id, name, tags, created_at. Required: prompt_version_id (must be a UUID, not the short prompt ID).
- **Helicone prompts rename** (Action). Rename an existing prompt in Helicone by updating its display name. Returns an empty 204 response on success. Required: prompt_id, name.
- **Helicone prompts get environment version** (Get). Get the prompt version assigned to a specific environment in Helicone (e.g., production, staging, development). Returns a version object including id, model, prompt_id, major_version, minor_version, commit_message, environment, created_at, and s3_url. Required: promptId, environment.
- **Helicone prompts get production version** (Get). Retrieve the currently designated production version of a prompt in Helicone. Returns: id, model, prompt_id, major_version, minor_version, commit_message, created_at, s3_url. Required: promptId.
- **Helicone prompts get version counts** (Get). Retrieve version statistics for a prompt in Helicone, including the total number of versions and major versions. Returns: totalVersions, majorVersions. Required: promptId.
- **Helicone prompts get version** (Get). Get a specific Helicone prompt version by its version id, including the full prompt body content. Returns: id, model, prompt_id, major_version, minor_version, commit_message, environment, created_at, s3_url. Required: promptVersionId.
- **Helicone prompts get versions** (Get). Get all versions of a specific Helicone prompt, optionally filtered by major version number. Returns an array of prompt version objects with id, model, prompt_id, major_version, minor_version, commit_message, created_at, and s3_url. Required: promptId.
- **Update a Helicone prompt by ID** (Update). Create a new version of an existing Helicone prompt with updated content, as either a major or minor version. Returns: id. Required: promptId, promptVersionId, newMajorVersion, commitMessage, promptBody.
- **Helicone prompts update environment** (Update). Set the environment for a specific Helicone prompt version. Environments can be production, staging, development, or any custom environment name. Returns null on successful update. Required: promptId, promptVersionId, environment.
- **List all Helicone prompts** (List). List Helicone prompts based on search criteria and tag filters. Returns an array of prompt objects with id, name, tags, and created_at. Required: search, tagsFilter.
- **Helicone prompts bulk update** (Update). Update tags for a Helicone prompt, replacing all existing tags with the new set provided. Returns the updated array of tag strings. Required: prompt_id, tags.
- **Helicone sessions submit feedback** (Action). Submit feedback (a boolean rating) for a Helicone session. Returns: data, error. Required: session_id, rating.
- **List all Helicone sessions** (List). Query Helicone sessions with filters, search, time-range filtering, and pagination. Returns: created_at, latest_request_created_at, session_id, session_name, total_cost, total_requests, prompt_tokens, completion_tokens, total_tokens, avg_latency, user_ids. Accepts a complex filter object with request/response-level and session-level conditions.
- **Helicone sessions list metrics** (List). Query Helicone session metrics with time-based aggregation and filtering. Returns: session_count, session_duration, session_cost, average. Accepts nameContains, timezoneDifference, useInterquartile, and a complex filter object.
- **List all Helicone users** (List). List user data in Helicone by user IDs. Returns: cost, user_id, completion_tokens, prompt_tokens, count. Required: userIds.
- **Helicone users list metrics** (List). List user metrics in Helicone with filters and pagination. Returns: hasUsers, count, users.
- **Helicone users list metrics overview** (List). Query user metrics overview in Helicone with interquartile and percentile-based filtering. Returns: user_cost, request_count. Required: useInterquartile, pSize, filter. pSize must be one of p50, p75, p95, p99, p99.9.
- **Create a Helicone webhook** (Create). Create a webhook in Helicone by specifying a destination URL, a configuration object, and whether to include enhanced data in payloads. Returns: data, error. Required: destination, config.