Connect Claude Code, Codex and ChatGPT to one memory over MCP
Set up Claude Code MCP, Codex and ChatGPT so the AI apps your team already pays for share one long-term memory: client history, decisions, house style. Three commands, one URL, and a clear line on where your prompts end up.
Quick answer
Run claude mcp add --transport http locul https://team-api.locul.ai/mcp, then /mcp in Claude Code to sign in. For Codex, run codex mcp add locul --url https://team-api.locul.ai/mcp and codex mcp login locul. For ChatGPT, switch on developer mode under Settings, Apps and connectors, Advanced settings, and create a connector with the same URL. All three then share one Locul memory. Your prompts go to Anthropic or OpenAI in the US.
Where do my prompts go with Claude Code, ChatGPT or Codex?
Your prompts go to Anthropic or OpenAI in the United States, together with any pieces of Locul memory the app pulls in to answer them.
Locul's memory itself stays where it lives: on your laptop with the desktop app, or in your team's Team Brain. MCP (the Model Context Protocol) lets the AI app ask Locul a question, such as "what did we agree with Brightwater Dental about VAT?", and Locul hands back only the matching notes. Those notes then travel with your prompt to the model.
US providers. Claude runs on Anthropic's infrastructure and ChatGPT and Codex run on OpenAI's. That is fine for code, public drafts and internal admin. For client files, HR cases or health data, use an EU or in-country provider, your own server or a local model instead.
How do I add an MCP server to Claude Code?
Run claude mcp add with the server's URL, then type /mcp inside Claude Code to sign in.
Add Locul as a remote server
In a terminal, from any folder:
bash$ claude mcp add --transport http locul https://team-api.locul.ai/mcpThat writes the server to your local scope. Add
--scope userto use it in every project, or--scope projectto share it with your repo through.mcp.json.Sign in
Start Claude Code, type
/mcp, pick locul and follow the browser sign-in. You can also runclaude mcp login loculfrom the terminal.claude code$ claude > /mcpCheck it works
Ask something only your memory knows. Claude Code calls Locul's search tool, reads the result and answers with the source.
"HTTP servers are the recommended option for connecting to remote MCP servers."
Claude Code docs, Connect Claude Code to tools via MCP
Which scope should I pick?
| Scope | Stored in | Who sees it | Use it when |
|---|---|---|---|
| local (default) | ~/.claude.json, under this project | Only you, only this project | You are trying Locul out |
| project | .mcp.json in the repo | Everyone who clones the repo | The whole team should use the same memory |
| user | ~/.claude.json | Only you, every project | Locul is your memory everywhere |
Only using the Locul desktop app? Open Settings, then AI apps, and press the Claude Code or Codex button. Locul writes the config for you and the memory is read straight from your laptop. The prompts still go to the US, as above.
How do I add an MCP server to Codex?
Run codex mcp add with --url, then codex mcp login to finish the OAuth sign-in.
$ codex mcp add locul --url https://team-api.locul.ai/mcp
$ codex mcp login loculCodex saves servers in ~/.codex/config.toml, or in .codex/config.toml inside a project you have marked as trusted. You can also write the entry by hand:
[mcp_servers.locul]
url = "https://team-api.locul.ai/mcp"In the Codex desktop app the same setting sits under Settings, MCP servers, Add server.
How do I connect ChatGPT to an MCP server?
Turn on developer mode in ChatGPT's settings, then create a custom connector that points at Locul's MCP URL.
Turn on developer mode
In ChatGPT on the web, open Settings, then Apps and connectors, then Advanced settings, and switch on Developer mode. On a Business or Enterprise workspace, only an admin or owner can do this.
Create the connector
Choose Create, give it the name Locul, paste
https://team-api.locul.ai/mcpas the server URL and pick OAuth as the sign-in method.Use it in a chat
Start a new chat, add Locul from the tools menu and ask your question. ChatGPT shows each tool call before it runs.
"ChatGPT web uses remote MCP tools through plugins."
OpenAI, Model Context Protocol in Codex and ChatGPT
Developer mode is labelled beta by OpenAI and the menu names move now and then. Which plans can use it also changes, so check OpenAI's help article if a setting is missing.
What does this look like on a real Friday?
You ask the AI app a question in plain words, and it checks Locul before it touches anything.
Here is the kind of message an agency owner actually types into Claude Code or Codex at the end of the month:
Claude Code calls Locul's search tool with "Brightwater Dental VAT setup fee", gets back Dana's call note if one exists, and quotes it before editing the script. If nothing matches, it stops, because you told it to. The prompt and Dana's note both travel to Anthropic, which is why this path suits internal admin better than a client's medical records.
When should a regulated firm avoid the frontier path?
Avoid it for anything that contains personal or client data you have promised to keep in your country.
| Task | Frontier path | Better path |
|---|---|---|
| Code, scripts, spreadsheets with no personal data | Good fit | Frontier is fine |
| Marketing copy, public web pages | Good fit | Frontier is fine |
| Client contracts, matter files, case notes | Check your client terms first | In your country or on-premise |
| HR, payroll, health or special category data | Usually not | On-premise or local |
You do not have to pick one. In Locul each bot has its own model, so the coding bot can use Claude while the HR bot runs on your office server.
How do frontier models compare with open models on office work?
We are measuring that now on real, messy office tasks, and the results will appear here.
Benchmark in progress
Forwarded email chains, broken spreadsheets and half-written contracts, scored by people who do this work. Frontier and open models, side by side. Coming soon.
Frequently asked questions
What is MCP in Claude Code?
MCP, the Model Context Protocol, is the open standard Claude Code uses to talk to outside tools and data. An MCP server such as Locul exposes tools like search and fetch, and Claude Code calls them when your prompt needs them.
Where does Claude Code store MCP servers?
Local and user scoped servers live in ~/.claude.json. Project scoped servers live in .mcp.json at the root of your repo, so they can be committed and shared.
Can ChatGPT use MCP servers?
Yes. With developer mode switched on, ChatGPT on the web can add a remote MCP server as a custom connector. OpenAI labels developer mode as beta, and on Business and Enterprise workspaces only admins and owners can turn it on.
Is my whole Locul memory sent to Anthropic or OpenAI?
No. The AI app only receives what Locul returns for each tool call, usually a handful of matching notes. Those notes do travel to the US with your prompt, so keep sensitive material on an in-country, on-premise or local model.
Can Claude Code and Codex share the same memory?
Yes. Point both at the same Locul MCP URL, or use the desktop app's AI apps buttons for both. Anything one app saves to Locul is there for the other the next time it asks.
Sources
Every command and quote on this page comes from these primary sources, checked on 24 October 2026.
Junaid builds Locul and sets up private AI for small regulated firms. He tests every command on this page on his own machines before it ships.