proxium

Connect agents

The MCP server

One URL for every client. You sign in with your proxium account.

Turn memory on first. In the console, open Settings > Memory of the project and choose on. While it is off, an agent can read nothing and save nothing.

https://proxium.tech/mcp

The server speaks Streamable HTTP. The first call opens the proxium sign-in in your browser (OAuth), and the agent then acts as you, in the projects you are a member of. It gives every client the same three things:

  1. Nine tools.proxium_recall_memory reads the project memory at the start of a task, proxium_search_memory and proxium_read_conversation find one thing, proxium_save_memory, proxium_correct_memory, proxium_rate_memory and proxium_forget_memory write back, and proxium_query_graph and proxium_list_projects read the rest.
  2. Instructions.The rules for when to read and when to write. Claude Code, ChatGPT, Gemini CLI and VS Code give them to the model with the tools.
  3. A prompt.proxium_load_memory starts a conversation with the project's knowledge page, its most trusted memories and its newest conversations.

Add it to your client

The same URL; the place it goes is the client’s.

Claude Code. Run the command, then /mcp in a session to sign in. The prompt is /mcp__proxium__proxium_load_memory.

claude mcp add --transport http proxium https://proxium.tech/mcp

Claude Desktop and claude.ai. Open Customize > Connectors, choose Add custom connector, paste the URL and choose Add. On a Team or Enterprise plan, an owner adds it under Organization settings > Connectors, and each member chooses Connect.

Codex (the CLI, the IDE extension and the ChatGPT desktop app). Add the server, then sign in. Codex shows a server’s tools to the model only after the model searches for them, so add the skills (3) or the project rule (4) too. Without them, a Codex session does not start with the memory.

codex mcp add proxium --url https://proxium.tech/mcp
codex mcp login proxium

ChatGPT. Turn on Developer mode under Settings > Security and login. Then create an app with the URL at chatgpt.com/plugins, and pick it from Developer mode in a chat.

Cursor. Add it to .cursor/mcp.json for one project, or to ~/.cursor/mcp.json for all of them. Cursor asks you to sign in.

{ "mcpServers": { "proxium": { "url": "https://proxium.tech/mcp" } } }

VS Code with GitHub Copilot. Add it to .vscode/mcp.json. A browser window opens to sign in on the first connection. The prompt is /proxium.proxium_load_memory.

{ "servers": { "proxium": { "type": "http", "url": "https://proxium.tech/mcp" } } }

Gemini CLI. The CLI signs you in when the server first asks, and /mcp auth proxium does it by hand. The prompt is /proxium_load_memory.

gemini mcp add --transport http proxium https://proxium.tech/mcp

Zed. Open Settings > AI > MCP Servers, then Add Server and Add Remote Server, or add it to the settings file. Zed asks you to sign in.

{ "context_servers": { "proxium": { "url": "https://proxium.tech/mcp" } } }

Windsurf (Devin Desktop). For the Devin agent, run the command, then devin mcp login proxium. For the Cascade agent, add { "mcpServers": { "proxium": { "serverUrl": "https://proxium.tech/mcp" } } } to its MCP config file.

devin mcp add proxium https://proxium.tech/mcp

Install the skills

They tell an agent when to read the memory and when to write.

Three Agent Skills (SKILL.md): proxium-memory, the loop; proxium-memory-recall, reading; proxium-memory-save, writing. The script puts them in ~/.agents/skills, which Codex, Gemini CLI, Cursor, GitHub Copilot, VS Code, OpenCode, goose and Amp read, and in ~/.claude/skills for Claude Code. With --project it writes them into the current repository instead, for the whole team.

curl -fsSL https://proxium.tech/skills/install.sh | sh
curl -fsSL https://proxium.tech/skills/install.sh | sh -s -- --project

Read the script first if you prefer: /skills/install.sh. The skills themselves are proxium-memory, proxium-memory-recall and proxium-memory-save; a folder with that name and the file in it is all a client needs.

Make it a project rule

For a client with no skills, or to be sure.

Put these lines in the file your client reads on every request: AGENTS.md (Codex, Cursor, GitHub Copilot, VS Code), GEMINI.md (Gemini CLI) or CLAUDE.md (Claude Code).

## Project memory
This project has a proxium memory (MCP server "proxium").
- At the start of every task, call proxium_recall_memory and use what it returns.
- Before you assume something about the project, search with proxium_search_memory.
- When the task settles a decision, a preference, a fact or a procedure a later
  session needs, save it with proxium_save_memory, one statement per call.
- Rate the memories you used with proxium_rate_memory, and correct the wrong ones.

Start each session with the memory

Automatic in Claude Code, one command elsewhere.
  1. Any client with prompts.Run the prompt proxium_load_memory at the start of a session: it puts the project memory into the conversation. Claude Code, VS Code and Gemini CLI show it as a slash command.
  2. Claude Code.The proxium-memory plugin bundles the server, the skills and a session-start hook. Set the plugin's api_key option to a virtual key of the project, and every new session starts with the project memory, with no command.

Connect one agent, and save the first decision it makes.

Start routing Every endpoint and default