Skip to main content

connecting edgeful to Claude, ChatGPT, and Grok with MCP

add the edgeful MCP connector to Claude, ChatGPT, or Grok, sign in with your edgeful account, and pull live report data in a normal chat - no API key needed

Written by Brad

edgeful data, inside the AI you already use

edgeful runs an MCP server — a connector that puts the platform's full report catalog inside your chat with Claude, ChatGPT, or Grok. you add it once, sign in with your edgeful account, and from then on you ask for data in plain english and the agent pulls the live numbers.

no API key to generate. no project folder. no code. setup takes about a minute. the API dashboard has a guided version too: pick chat apps, then Claude, ChatGPT, or Grok, and follow the short setup video.

the server URL, if you want it up front: https://api.edgeful.com/mcp

MCP or the API — which one do you want?

edgeful has two ways to get report data to an AI tool, and they're for different jobs:

  1. MCP (this article) — you want to ask questions. you're in a normal chat with Claude, ChatGPT, or Grok, you want real numbers instead of guesses, and you don't want to install or maintain anything. nothing lives on your machine

  2. the API + a coding agent — you want to build something. dashboards you regenerate every morning, scripts you save, files you keep. that path needs an API key and a workspace, and it's covered in setting up the edgeful API with VS Code + Claude Code

same data behind both. you can run both at once — plenty of people connect MCP for quick questions and keep the API project around for dashboards.

what you'll need

  1. an account on the chat client you use. in Claude, custom connectors work on Free, Pro, Max, Team, and Enterprise — Free accounts are limited to one connector total. in ChatGPT, custom connectors sit behind developer mode, which is on Plus, Pro, Business, Enterprise, and Edu but not the free plan. Grok takes the connector from a GrokBot chat or from the connectors page on the web

  2. an active edgeful plan — essentials, pro, and all-access all include API access, and MCP rides on the same entitlements. signing in works from any account, but a tool call from an account with no active plan comes back with a missing_entitlement error

that's the whole list. you do not need an API key for this.

connecting it to Claude

works in the Claude desktop app and on the web.

  1. in Claude, open Customize → Connectors from the sidebar, click Add, then Add custom connector. older versions of the desktop app have this under Settings → Connectors instead, which is where the video below starts

  2. name it edgeful and set the URL to https://api.edgeful.com/mcp, then save

  3. click Connect. a browser window opens an edgeful sign-in and consent screen — not a bare API key prompt. sign in with your edgeful account and click Allow

  4. back in Claude, the edgeful tools are live in your conversation

that's it, and you only do it once. the connection stays put across new chats.

one more setting worth changing: tool permissions. open the edgeful connector under Customize → Connectors and you'll see two groups, both set to needs approval:

  • read-only tools — the tools that pull reports and list your custom reports. set these to always allow so Claude doesn't stop to ask every time it reads data. they can't change anything

  • write/delete tools — the tools that change things in your account: building, revising, running, saving and deleting custom reports, plus your bookmarks, watchlists, account defaults, custom sessions, report customizations and templates. leave this group on needs approval unless you're happy for Claude to make those changes without asking

watch the full Claude setup and a first question in about 3 minutes:

connecting it to ChatGPT

ChatGPT takes the connector through developer mode. it's a paid-plan feature — Plus, Pro, Business, Enterprise, and Edu — and on Business and Enterprise an admin may have to allow custom connectors before the option shows up.

  1. click your profile and open Settings → Plugins. OpenAI renamed connectors to plugins in July 2026, so older tutorials show the old label

  2. scroll to advanced settings and turn developer mode on

  3. back on the plugins screen, add a custom connector. name it edgeful and paste https://api.edgeful.com/mcp

  4. choose the sign-in option, sign in with your edgeful account in the browser tab that opens, and click Allow

  5. in a new chat, enable the edgeful connector from the tools menu and ask your first question

this is ChatGPT itself — the website or the app. if you'd rather have the connector inside OpenAI's coding agent, that's Codex, covered further down.

connecting it to Grok

GrokBot runs on its own cloud computer, so you add the connector by telling the bot to do it:

  1. in a GrokBot chat, type: add a custom MCP server called edgeful at https://api.edgeful.com/mcp

  2. the bot shows you the name and URL to confirm. select add it

  3. a connect card appears in the chat. click authorize, sign in with your edgeful account in the browser tab that opens, and approve access

  4. type @ in the chat and pick edgeful to attach the server to your task, then ask a question

using Grok on the web instead of the bot? go to grok.com/connectors, choose new connector → custom, and use the same URL and sign-in.

one thing that trips people up: because GrokBot lives in the cloud, a localhost URL never works with it. the edgeful server is remote, so that limit doesn't apply here.

your first question

start with something simple so you can watch it work:

what are the top reports for ES right now?

you'll see the agent make a tool call before it answers. that's your confirmation — it's pulling from edgeful, not guessing from memory. if it answers without calling a tool, just say use the edgeful tools for this and it'll go get the real data. after the first tool call in a chat it keeps using them on its own.

from there, ask for whatever you'd normally go look up:

  1. pull the ORB report for NQ, New York session, over the last year

  2. compare ES and NQ on gap fill and stack them side by side

  3. what does the IB report say for CL this month?

you never paste screenshots of reports. the connector pulls the live data straight from edgeful — it's working with the actual numbers, not an image of them.

one habit worth building: before it runs a report you haven't used, ask it to describe the report first. the description spells out what the report counts and which session and settings it defaults to, so the number it quotes is the one you meant.

what the connector actually gives your agent

you don't call these tools by name. you ask in plain English and the agent picks the right one, but it helps to know what's behind the curtain. 4 tools read report data:

  1. list_allowed_report_endpoints — which reports your plan can reach

  2. describe_report_endpoint — what a given report returns and which settings it accepts (ticker, session, lookback, variant)

  3. call_report_endpoint — the one that actually pulls the data

  4. get_discovery_scan — pulls a discovery scan

it's the same catalog the REST API exposes, so anything you can see in the platform, you can ask for here.

8 more handle your custom reports (see the next section):

  1. build_custom_report — builds a new report from your description

  2. revise_custom_report — changes a draft in the same chat

  3. save_custom_report — saves the reviewed report to your account

  4. run_custom_report — runs a saved report or a draft on a ticker, session and date range

  5. get_custom_report_run — reads a run by its id, 100 rows at a time

  6. list_custom_reports — lists your saved reports

  7. get_custom_report_latest_result — reads the last stored result without rerunning

  8. delete_custom_report — deletes a saved report

12 more manage your edgeful settings (see managing your settings from Claude below):

  1. get_account_settings — reads your defaults, custom sessions, bookmarks, watchlists and templates

  2. list_report_catalog — lists the report pages you can bookmark, set as a default or save a template on

  3. describe_report_customizations — shows the customize options on a report page

  4. update_report_customizations — changes or resets a report page's customizations

  5. update_bookmarks — adds and removes report bookmarks

  6. update_watchlist — adds and removes tickers on a watchlist

  7. update_profile_defaults — changes your default market, ticker, session, date range or report

  8. save_custom_session — creates or updates a custom session

  9. delete_custom_session — deletes a custom session

  10. list_templates — lists your templates, plus the built-in ones (read-only)

  11. save_template — creates or updates a report, What's in Play or screener template

  12. delete_template — deletes a template

and 1 sends feedback to edgeful:

  1. submit_feedback — reports a bug, a feature request or something confusing, for the edgeful team to review

so if something in edgeful gets in your way, you can tell Claude without leaving the chat. say what happened and what you expected, and it sends it as a bug, a feature request or confusing behavior. the edgeful team reviews every submission. sending feedback follows the same write switch as custom reports and settings, and it's capped at 10 a minute and 100 an hour. leave out passwords, API keys and private trading details.

building custom reports from Claude

custom reports are for the questions edgeful doesn't have a report for yet. with the connector you can build one from a Claude chat instead of opening edgeful AI.

  1. check that Claude can write to your account. in the API dashboard, under let Claude or your code write to your account, connected Claude covers every Claude connection signed in to your edgeful account, and this API key covers anything that uses your API key: a client like Cursor (or another mcp-remote setup), or your own code calling the API. both are on unless you've turned them off

  2. in Claude, describe the report in plain English: the ticker, the time window with its timezone, and exactly what to count

  3. check the sample against a few days on your own chart, ask for changes if anything is off, then tell Claude to save it

builds and revisions use your edgeful AI allowance. running a saved report doesn't. saved reports show up with your other custom reports in edgeful AI, marked built via API, and you can rerun them on another ticker or date range from Claude or from edgeful AI.

a few limits worth knowing: Claude reads a result 100 rows at a time, so ask it to keep going if you want every row. runs are capped at 60 a minute and 600 an hour, and builds, revisions, saves and deletes at 10 a minute and 100 an hour. both caps are shared across all your connections and keys, including your own code. turning the switch off stops writes, and reading reports keeps working.

for writing a good description and checking the sample, see how to build custom reports with edgeful AI. André walks through a full example in London session time in ET (blog).

managing your settings from Claude

the same switch lets Claude change your edgeful setup from a chat. ask it to bookmark a report, add tickers to a watchlist, change your default ticker or session, save a custom session, change a report's customizations, or save a report, What's in Play or screener template.

changes show up in edgeful after you reload the page. reload any edgeful tab you had open before Claude made the change, too: an older tab can overwrite Claude's bookmark and watchlist edits the next time it saves. a template Claude saves appears in that page's template picker (the report sidebar, What's in Play or the screener), and you apply it from there.

changing settings doesn't use your edgeful AI allowance. settings changes are capped at 60 a minute and 600 an hour, shared with anything else using your account or keys. if you'd rather Claude only read, turn off connected Claude (or this API key) and it can still read your reports and settings without changing anything.

your own code can make the same changes with your API key, and list, run, read, save and delete your saved custom reports. building and revising a report stay with Claude and edgeful AI. the routes are in manage your settings with the API in the developer docs.

using Codex (OpenAI's coding agent)

Codex is the coding-agent version of ChatGPT — it lives in the ChatGPT desktop app and as a CLI. it connects the same way — sign-in, no key:

  1. in the ChatGPT desktop app, open Settings → Plugins (under Integrations), then Add → Add MCP server. in the connect to a custom MCP form, name it edgeful, set the type to Streamable HTTP, and use https://api.edgeful.com/mcp. leave the bearer token and headers fields empty. save

  2. the server shows up flagged as needing OAuth. click Authenticate, sign in with your edgeful account in the browser tab that opens, and click Allow

  3. restart if Codex asks you to. typing /mcp in the composer lists your active servers so you can confirm edgeful is there

the app and the CLI share the same config file — ~/.codex/config.toml — so a server you add in the app shows up in the CLI too. if the CLI asks you to reauth, run codex mcp login edgeful.

clients that can't sign in yet (Cursor, older setups)

some MCP clients don't run the sign-in handshake and only accept a static token. for those you keep using an edgeful API key, pointed at the same server through mcp-remote.

generate a key in your API dashboard under API Keys — see getting your first API key for the full walkthrough — then add this to your client's MCP config:

{
  "mcpServers": {
    "edgeful": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.edgeful.com/mcp",
        "--header",
        "Authorization: Bearer ${EDGEFUL_API_KEY}"
      ]
    }
  }
}

then set EDGEFUL_API_KEY in your environment, so the key itself never sits in the config file.

never paste the key directly into a config file you might share or commit.

between the setups above, this covers Claude desktop and web, ChatGPT, Grok, Codex in the ChatGPT app or the CLI, Cursor, and any other MCP client that speaks Streamable HTTP. if your tool isn't named here, it's one of these two paths.

and the coding agents — Claude Code, Codex, Cursor — don't need the connector at all if you're building something. they can read the API docs and call the API directly with a key in a .env file. that's the build path in setting up the edgeful API with VS Code + Claude Code.

plan access

MCP is gated exactly like the REST API — same entitlements, same tier limits. every edgeful plan includes API access, so essentials, pro, and all-access all work. what changes between them is which reports and tickers the connector can reach and how far back it can look: essentials is limited to a handful of reports and four test tickers, while pro and all-access open up the full catalog. so if the agent comes back saying a report or ticker isn't available to you, that's your tier talking, not a broken connector — the breakdown is in rate limits and tier differences.

separately, sign-in itself works from any account, so the connector can look successfully connected right up until the first tool call comes back with missing_entitlement. that one means there's no active plan on the account — usually a lapsed or cancelled subscription. it's a billing question, not a setup problem: check your plan in your edgeful dashboard or reach out in the in-app chat.

what the connector doesn't do

a few things people expect from it that it isn't built for:

  • it doesn't push alerts. the connector only answers when you ask. if you want something that reacts as the session moves, the API has live what's in play and screener streams (pro and all access) that push updates, and a coding agent can build on those

  • it doesn't predict. every number that comes back is how often something happened in the window you asked about

  • it doesn't touch your billing, plan or login. the report tools are read-only. what it can change is your custom reports and your edgeful setup (bookmarks, watchlists, defaults, custom sessions, report customizations and templates), plus sending feedback to edgeful, and you can switch that off in the API dashboard

and one thing to keep separate: edgeful AI, the assistant inside the platform, is a different product from the connector. edgeful AI answers questions in the app; the connector brings the same reports into whatever AI you already use. see edgeful AI.

if something doesn't line up

  1. no browser window opened when you clicked Connect — check your pop-up blocker, then confirm the URL is exactly https://api.edgeful.com/mcp with no trailing slash or stray characters

  2. the tools don't show up in a chat — fully restart the client so it picks up the new connector, and start a fresh conversation

  3. developer mode is missing in ChatGPT — it isn't on the free plan, and on Business and Enterprise an admin may have to allow custom connectors first

  4. ChatGPT shows edgeful connected but won't use it — make sure the connector is enabled from the tools menu in that chat

  5. the agent answers without pulling data — tell it to use the edgeful tools. naming the ticker, session, and lookback in your question also nudges it toward a real call

  6. missing_entitlement — see the plan access section above

  7. other error codes — the API and MCP share the same backend, so API errors and troubleshooting applies here too

one note on scope: this article covers the connector. if you're setting up the key-based route in VS Code and a tool is what's stuck, see edgeful API setup troubleshooting: common errors instead.

where to take it from here

MCP is the fastest way to get real edgeful numbers into a conversation, and for a lot of people it's all they ever need. when you start wanting the output saved — a dashboard that regenerates every morning, cross-report mashups you keep coming back to — that's when the API path earns its keep. see building custom dashboards with the edgeful API + Claude Code.

the setup docs live at edgeful.com/docs/getting-started/set-up-mcp, and the full endpoint catalog is in the developer docs.

André's walkthrough of all six clients — with short setup videos for Claude, ChatGPT, and Grok — is on the blog: how to connect the edgeful MCP to Claude, ChatGPT, and GrokBot (blog).

related articles

Did this answer your question?