← Back to Documentation

DeerDawn MCP Server

Connect any AI tool to your DeerDawn workspace with one remote URL and a browser sign-in — no API keys to copy. Claude Code, Cursor, Codex, Claude.ai, and ChatGPT all use the same hosted endpoint; coding tools can optionally run a local npx process for repo-file writes.

Local tools

Claude Code, Cursor, Codex, Claude Desktop, and ChatGPT Desktop all run the MCP server as a local npx -y @deerdawn/mcp-server@latest process. No API key needed in the config — a browser sign-in happens on first use.

Claude Code

Run in your terminal — Claude Code has no config file

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

Cursor

File: .cursor/mcp.json (project) or ~/.cursor/mcp.json (global)

{
  "mcpServers": {
    "deerdawn": {
      "command": "npx",
      "args": ["-y", "@deerdawn/mcp-server@latest"],
      "env": {
        "DEERDAWN_API_URL": "https://api.deerdawn.com"
      }
    }
  }
}

Codex

File: ~/.codex/config.toml

[mcp_servers.deerdawn]
command = "npx"
args = ["-y", "@deerdawn/mcp-server@latest"]
startup_timeout_sec = 120

[mcp_servers.deerdawn.env]
DEERDAWN_API_URL = "https://api.deerdawn.com"

Claude Desktop

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "deerdawn": {
      "command": "npx",
      "args": ["-y", "@deerdawn/mcp-server@latest"],
      "env": {
        "DEERDAWN_API_URL": "https://api.deerdawn.com"
      }
    }
  }
}

Restart Claude Desktop after saving. The server will prompt you to sign in via browser on first use.

ChatGPT Desktop

  • macOS: ~/Library/Application Support/com.openai.chat/mcp_servers.json
  • Windows: %APPDATA%\ChatGPT\mcp_servers.json
{
  "mcpServers": {
    "deerdawn": {
      "command": "npx",
      "args": ["-y", "@deerdawn/mcp-server@latest"],
      "env": {
        "DEERDAWN_API_URL": "https://api.deerdawn.com"
      }
    }
  }
}

Restart ChatGPT after saving. Sign in via browser when prompted.

Verifying a local connection: Ask your tool to run start_session or "Check DeerDawn auth status." If authentication is needed, get_auth_status returns a browser sign-in URL. Once signed in, list_projects confirms the connection is live.

Web tools — Claude.ai and ChatGPT.com

Claude.ai and ChatGPT.com support remote MCP servers over HTTPS. DeerDawn provides a hosted endpoint — no local install needed. Claude.ai authenticates via browser OAuth, the same flow as local tools.

Step 1 — Add the remote MCP

Claude.ai

Open claude.ai
Customize
top-right menu
Connectors
+ (Add Connector)
Add custom connector
Enter MCP server URL
https://api.deerdawn.com/api/v1/mcp
Click Connect
Authorize DeerDawn in browser
DeerDawn tools available ✓
Open chatgpt.com
Apps
Settings
Developer mode already enabled?
No
Advanced settings
Developer mode ON
Yes
Skip
continue below ↓
Create App
Enter MCP server URL
https://api.deerdawn.com/api/v1/mcp
(Optional) Import logo
use the SVG downloaded above
Connect / Save
DeerDawn tools available ✓

After connecting

Your DeerDawn connector appears at Apps → DeerDawn. Review available actions, check connection status, and manage or revoke access from there.

Verifying the remote connection: Ask Claude or ChatGPT to call list_projects. It should return your DeerDawn workspaces. If you get an auth error, try disconnecting and reconnecting to re-trigger the OAuth flow.

Step 2 — Add the system prompt

Just having the tool available isn't enough — the model won't call start_session automatically unless you tell it to. Add the snippet below to your system prompt so DeerDawn context loads at the start of every conversation without you asking.

Claude.ai

Add to a Project (Settings → Projects → Project instructions) so it applies to every conversation in that project. Or add to your personal instructions if you want it everywhere.

At the start of every conversation, silently call start_session from DeerDawn before responding. Use the returned context (current task, decisions, stack) to answer without asking me to re-explain the project.

After any of the following, call update_context with a 1-2 sentence summary:
- A decision is made
- The task or status changes
- Something meaningful is built or fixed
- I confirm an approach ("yes", "good", "do it", "ship it")

Do not mention that you are calling these tools unless I ask.

ChatGPT.com

Add to Settings → Custom instructions → What would you like ChatGPT to know? for all chats, or paste it into the system prompt of a Custom GPT if you use one for coding work.

At the start of every conversation, call start_session from DeerDawn before responding. Use the returned context to orient yourself — current task, recent decisions, open questions — without asking me to re-explain.

After any decision, task change, or meaningful progress, call update_context with a brief summary of what changed and why.

Call these tools silently. Do not narrate the tool calls unless I ask.

Why this works: Both Claude.ai and ChatGPT treat system prompt instructions as high-priority directives, not suggestions. With this in place, start_session runs before your first message in every conversation — the same guarantee the local tools get from their SessionStart hooks.

What the MCP server does

Context retrieval

  • Current task and status
  • Recent decisions
  • Open questions
  • Goals and tech stack

Context updates

  • Write progress summaries
  • Record decisions automatically
  • Mark goals complete
  • Resolve open questions

Session management

  • Start a session and orient instantly
  • Import local CLAUDE.md / AGENTS.md
  • Switch between projects
  • Search across context

Works across tools

  • Claude Code, Cursor, Codex
  • Claude Desktop
  • Claude.ai web
  • ChatGPT Desktop + ChatGPT.com

Automatic context cleanup

Completed tasks, resolved questions, and finished goals are automatically removed from your context graph as you work. You don't need to manually clean up — DeerDawn infers completion from what you tell it.

Completed goals are removed
When you tell DeerDawn a goal shipped — "shipped the auth flow" — that goal node disappears from your context map automatically.
Resolved questions are removed
When an open question gets answered — "decided to use Postgres" — the question node is removed and a decision node takes its place.
Old tasks are replaced, not stacked
When the current task changes, the previous task node is removed. Your graph shows what's active now, not a graveyard of past work.
Example — tell DeerDawn something is done:
Update DeerDawn: shipped the onboarding flow goal, decided to drop Redis queue in favour of Postgres polling
This removes the "onboarding flow" goal node, removes the "Redis vs Postgres" question node if it existed, and adds a decision node — all automatically on the next update_context call.

Dashboard: You can also resolve questions manually in the Context Graph — click any open question node and pick or type an answer. The question is removed and converted to a decision immediately.

Usage examples

Start a session

Ask your AI tool:
Start a DeerDawn session and tell me what I was working on
Calls start_session — returns your active project context so you can resume without re-explaining the project.

Save a decision

Ask your AI tool:
Update DeerDawn: we decided to use Postgres for the events table instead of DynamoDB
Calls update_context — the decision is stored and will appear in every future session across all your connected tools automatically.

Check what's open

Ask your AI tool:
What open questions and goals do I have in DeerDawn?
Calls get_context with the questions and goals sections.

Security & privacy

DeerDawn is designed for professional context — tasks, decisions, and open questions that are often sensitive. Here's what we do to keep that data safe and give you control over it.

Encryption at rest

All context data is encrypted at rest using AES-256 and in transit using TLS 1.3+. DeerDawn stores only structured context deltas, never your raw conversations.

Full access log

Every time your context is read or written — by you, an AI tool, or the MCP server — an audit entry is created. View it anytime in the dashboard under Security.

Data exclusions

Mark specific context fields (open questions, recent decisions, etc.) as excluded in your preferences. Excluded fields are never written to the database, even if extracted.

TTL / auto-expiry

Set a TTL on your context data (e.g. 30 days). After that window, projects that haven't been updated are automatically purged.

Export your data

Download everything DeerDawn has stored — projects, graph nodes, and your full audit log — as a single JSON file from the Security page in the dashboard.

Verified deletion

Delete all your data with one click. The deletion is immediate and returns a signed confirmation token as proof it happened. No waiting period, no backups retained.

What DeerDawn stores
  • Current task, status, goals, open questions, recent decisions
  • Tech stack, file paths, and project metadata you've shared
  • Conversation excerpts used as evidence for extracted context
  • Secrets, passwords, API keys, or payment data — never stored
  • Full conversation transcripts — only structured extractions
For full security posture details, see the Security page. For data controls, visit the Security tab in your dashboard.

Configuration options

VariableRequiredDefaultDescription
DEERDAWN_API_URLYeshttps://api.deerdawn.comDeerDawn API base URL. Local tools only.
DEERDAWN_ENVIRONMENTNoproductionEnvironment override (sandbox or production)
DEERDAWN_CACHE_ENABLEDNotrueEnable response caching. Local tools only.

Local tools authenticate via browser OAuth — no API key in the config. Web tools (Claude.ai, ChatGPT.com) also authenticate via browser OAuth — no API key needed.

Resources

@deerdawn/mcp-server on npm

Connect DeerDawn in under 5 minutes

Need Help?

Join our community or reach out to our support team