Self-serve · MCP

Your context network,
in supported MCP clients.

DeerDawn MCP briefs Claude Code, Codex, and Cursor on all your projects, decisions, open questions, and working style, so every new session starts caught up, never cold. One local MCP install, one sign-in, same brief in every tool.

On Claude.ai? Connect via remote MCP →

Set up in 2 steps

1
Paste prompt 1, restart
Copy the first prompt below and paste it into your AI tool. It writes the config file for you and tells you to restart. One restart is required. MCP servers load at startup.
2
Paste prompt 2, done
After restarting, paste the second prompt. It signs you in, wires up your context file so DeerDawn loads every session, and imports your existing projects. Takes about 30 seconds.

Why this page exists

Most people land here after they realize the real problem is not prompt quality but context continuity. If ChatGPT, Claude, Codex, or Cursor keeps forgetting your project state, the practical fix is shared AI memory, a brief every new session reads, not more copy-paste.

how to share context between AI tools →How to make ChatGPT remember context →Claude memory between chats →

Let your AI set it up

Copy this and paste it into Claude, ChatGPT, or any AI tool. It will add the config to the right file for you.

1Paste in Claude Code before restart: sets up the config file
Please set up DeerDawn MCP for me.

Claude Code registers MCP servers through its CLI. It does NOT read a ~/.claude/mcp.json file. Run this command in my terminal:

claude mcp add -s user deerdawn -e DEERDAWN_API_URL=https://api.deerdawn.com -e DEERDAWN_SURFACE_ID=claude_code -- npx -y @deerdawn/mcp-server@latest

Confirm the command succeeded, then tell me to restart Claude Code. After I restart I'll paste the second prompt.
restart Claude Code, then
2Paste after restart: signs in, configures, imports context
DeerDawn MCP is now loaded. Before installing anything, check whether DeerDawn is already connected. Don't assume a fresh setup.

STEP 0: Check for an existing connection (always do this first):
Call start_session (or deerdawn-start_session if your host prefixes tool names).
- If it returns project or workspace context with no auth error, DeerDawn is ALREADY CONNECTED. Tell me "DeerDawn is already installed and ready" and STOP. Skip every step below, and don't request a sign-in URL.
- If it says I'm not authenticated, continue to step 1.
- If start_session isn't available on this host at all, DeerDawn isn't connected here, so continue to step 1.

1. Call get_auth_status (or deerdawn-get_auth_status). If I'm not authenticated, give me the sign-in URL. I'll open it, connect my account, and tell you when done.

2. Call start_session again. Now that I'm signed in, the local DeerDawn server writes my brief into this repo and wires it into CLAUDE.md, so it loads automatically at the start of every session.

3. Call manage_memory with action "import_local" and no confirmed flag first, to preview any CLAUDE.md, AGENTS.md, or .cursorrules in this repo. (The local server only. If the action is refused, skip steps 3 and 4.)

4. Then call manage_memory with action "import_local" and confirmed=true to complete the import.

When you're done, confirm what was set up and tell me DeerDawn is ready.

Or install manually

Choose your tool and paste the config. Node.js 18+ required.

Run in your terminal
claude mcp add -s user deerdawn -e DEERDAWN_API_URL=https://api.deerdawn.com -e DEERDAWN_SURFACE_ID=claude_code -- npx -y @deerdawn/mcp-server@latest

After pasting, restart your tool. On first use, a browser tab will open to sign in.

What your AI tools can access

start_session

Opens every session with the project's current state — task, decisions, open questions, known paths.

recall

Look up a stored project, its brief, the architecture you have recorded, or a known file path.

manage_memory

Write back what the session learned: a decision, a task change, shipped work, a file path, a debugging finding.

manage_tasks

Read the task board, and add, move, or complete cards on it.

manage_projects

Create, select, rename, move, archive, or restore a project.

build_subagent_brief

Produce a short, task-scoped brief to hand a delegated worker.

Related availability

More ways context flows in

Gemini
Remote MCP support for Gemini arrives once Google enables third-party MCP endpoints.
VS Code
MCP support in VS Code Copilot agent mode is tracked. Config path will mirror Cursor when available.

Start for free

Create a free account and set up DeerDawn MCP in minutes.