INTEGRATION GUIDE
MCP for D&D: Connect Campaign Memory to ChatGPT and Claude
Model Context Protocol lets an AI assistant work with your campaign record instead of relying on whatever fits in one chat window.
Published October 19, 2026 · 5 min read· Reviewed October 18, 2026
By Greg H. & Greg W., co-founders of Archivist
← Back to GM GuidesThe Short Answer
MCP (Model Context Protocol) lets ChatGPT, Claude, Cursor, Windsurf, and other AI apps read your Archivist campaign directly. Connect the Archivist server, sign in with your Archivist account, and ask about sessions, quests, characters, locations, factions, items, or journals. Archivist is listed in Anthropic's Connectors Directory for Claude and has an official ChatGPT app.
Why MCP Is Useful for a D&D Campaign
A normal chat only knows what you paste into it. A long campaign outgrows that fast. With MCP, Archivist stays the source of truth and your assistant asks it for the specific records a task needs.
That makes questions like these practical:
- What changed in the last three sessions?
- Which open quests involve the Thieves' Guild?
- When did the party first meet this NPC?
- Which threads have gone quiet?
- Create a journal from these prep notes, if write access is on.
The Model Context Protocol is an open standard, so the same connection works across assistants. Archivist's server exposes more than 80 campaign tools through it, some that read and some that write.
Workflow 1: Prep Your Next Session in One Prompt
The Archivist Cookbook has a copy-paste prompt for this. Once Archivist is connected (setup is below), the assistant lists your campaigns and asks which one you mean. It then reads your three most recent sessions plus your quests and characters, and builds a prep outline in a canvas or artifact you can keep open at the table.
The outline has five sections:
- Open threads: every unresolved thread and where it started. Threads untouched for two or more sessions are flagged as going cold.
- NPCs to bring back: characters with unfinished business, one line each on why now.
- Hooks: three openings that pay off choices your party already made.
- Likely questions: what players may ask about earlier events, answered from the record.
- Forgotten loose ends: anything that looks deliberate but never resolved.
The prompt tells the assistant to use only your campaign data, label anything inferred as a suggestion, and never invent NPCs, locations, or events.
Workflow 2: Automate the Post-Session Routine
Tom Herrmann has run the same campaign since 2019. After every session he exported notes to Notion, made artwork, posted summaries, and prepped recap audio. Each step needed the same campaign data, and the Notion export alone took another 15 minutes of copying and pasting. So he built the Archivist MCP server himself.
I can tell Cursor to process the latest session, wait about fifteen minutes, and come back to a completed Notion page, session artwork for Canva, social media images, and NotebookLM audio files. One click and roughly 90% of my post-session workflow is done.
Tom said that in How One Archivist User Automated 90% of His Post-Session Workflow. The Cookbook also has recipes for a podcast-style session recap and a campaign arc infographic.
More Things to Ask Your Campaign
- Session catch-up: ask for events, decisions, and open threads from a date range, then ask for the source of any detail the group disputes.
- Continuity check: compare a draft scene against known locations, relationships, and past rulings. Treat conflicts as questions for you, not automatic corrections.
- Journal and quest updates: with write access on, turn approved notes into a journal entry or update an existing quest. Review the target before it runs.
- Anything outside Archivist: bring campaign context into a coding editor, writing tool, or image assistant. Archivist Nexus helps Claude Desktop users connect several MCP servers at once.
How to Connect Archivist MCP to ChatGPT, Claude, Cursor, or Windsurf
You need a paid Archivist plan. Any tier works. Then pick the path for your assistant:
- Claude: add Archivist from Anthropic's Connectors Directory.
- ChatGPT: install the official Archivist app.
- Claude Desktop, Cursor, or Windsurf: add the server URL https://mcp.myarchivist.ai/mcp. Archivist Nexus sets up Claude Desktop for you.
Then:
- Sign in with your Archivist account when the window opens.
- Approve the access the connector asks for.
- Start with a read-only question, such as "What happened in our last session?"
- Turn on write access later, only if you want the assistant to edit records.
Two terms you may see: the sign-in window is OAuth, the standard way to approve access without pasting a password into chat. Streamable HTTP is the connection type some clients ask for. Choose it and move on.
Client screens change often. For exact steps, use the current Archivist MCP setup guide. The Archivist MCP page lists every connection path.
Read Access, Write Access, and Safety
Read access lets the assistant look up campaign information. Write access lets it create or update records such as characters, quests, and journals. Writes need a separate permission, called agent_write, which you grant only when you want edits.
Start read-only. Before you approve a write, ask the assistant to show the proposed change, name the record it will touch, and say which source supports it. Keep private GM material out of player-facing prompts. Check the server URL and publisher before you connect, sign in through OAuth instead of pasting credentials into chat, and remove connections you no longer use from your Archivist profile.
MCP does not make every answer correct. It gives the model better access to the source. The assistant can still misread a record, propose a bad edit, or follow instructions hidden in connected content, so review anything that changes your campaign.
Where Discord MCP Fits
Archivist runs a second, separate server for Discord. Campaign MCP, covered here, reads and updates your campaign. Discord MCP lets your assistant search and post in Discord through a bot you create. The two use different URLs and different permissions, so you can connect either one or both.
For setup and examples, read How to Use a Discord MCP Server for TTRPG Campaign Management or the Discord MCP overview.
Frequently Asked Questions
What does MCP stand for in AI tools?
MCP stands for Model Context Protocol, an open standard that lets an AI app discover and call tools exposed by an external server.
Do I need a paid plan to use Archivist MCP?
Yes. Archivist MCP is included with every paid plan and is not available on the free trial.
How do I connect ChatGPT to my D&D campaign?
Install the official Archivist app in ChatGPT, sign in with your Archivist account, and ask about your sessions, quests, or characters. A paid Archivist plan is required.
How do I connect Claude to my D&D campaign?
Add Archivist from Anthropic's Connectors Directory, sign in with your Archivist account, and start a new chat about your campaign. A paid Archivist plan is required.
Does Archivist MCP work with ChatGPT and Claude?
Yes. Archivist is listed in Anthropic's Connectors Directory for Claude and has an official ChatGPT app. Cursor, Windsurf, and other clients that support hosted MCP servers with OAuth can connect with the server URL.
Can an MCP client edit my D&D campaign?
Only when you approve write access. Archivist keeps its write tools behind the agent_write permission, so start read-only if you only need questions and summaries.
What is the difference between Archivist Campaign MCP and Discord MCP?
Campaign MCP works with your campaign records inside Archivist. Discord MCP searches and posts in Discord through a bot you create. They use separate URLs and permissions, and you can use either or both.
Is MCP the same as uploading a campaign document?
No. An upload is a static copy inside one conversation. MCP lets the assistant request current records from your campaign when it needs them.
Connect the Campaign You Already Have
On any paid plan, start with read-only access, ask one question about a past session, and check the source before you expand the workflow.