Description
Rafter is a Model Context Protocol (MCP) server, built by a team in Bergen, Norway, that gives Claude, ChatGPT, Cursor, Claude Code and other MCP-compatible AI tools one shared memory for a team. Instead of every person's AI relearning how the company does things, Rafter stores the skills, decisions and workflows a team has already worked out, so any teammate's AI can look them up and reuse them.
Read More
Setup has two steps. First, a user adds Rafter as a custom MCP connector using its hosted endpoint inside Claude, ChatGPT, Cursor, Claude Code, Codex, OpenCode or Windsurf, which opens a sign-in window; Claude Code and Cursor use a CLI command or a config file entry instead. Second, the user pastes a short instruction into their AI tool's system or project instructions telling it to search Rafter's memories before answering and save anything durable back with an add_memory call. From then on, the AI checks Rafter automatically as part of normal use, and one person can also hand a teammate's tool a setup URL so it picks up a skill directly.
Rafter organizes what it stores as a knowledge graph rather than a flat memory list: it maps how documents, decisions and skills connect to each other, so a search follows related context instead of matching only keywords. On paid plans, it also connects to the tools teams already use for source documents, including Jira, Confluence, SharePoint and Drive, keeping that connected search current as the underlying documents change, and it keeps version history with rollback for stored skills and memories.
Rafter is aimed at teams who want their various AI assistants to share institutional knowledge instead of each starting from zero, particularly teams already working across multiple AI coding and chat tools. Pricing is per user per month: a Free plan covers up to three members with memory, skills, agents and hosted MCP access; a Team plan is $29 per user per month for up to 50 members, adding tool connectors, connected search, version history and cross-team sharing with a 14-day free trial; and a Business plan is $99 per user per month with unlimited members, SSO, audit log export, priority support and a signed DPA for organizations with formal IT and security requirements.
- CategoryAI Agent Builders
- Pricing ModelFreemium
- DeploymentCloud
- PlatformsWeb, Desktop
Rafter's Core Features
MCP server that gives Claude, ChatGPT, Cursor, Claude Code and other MCP clients one shared team memory
Knowledge graph links documents, decisions and skills so search follows connected context, not just keywords
add_memory tool lets any connected AI save durable context back to the shared store
A shared setup URL lets one teammate hand another AI tool a ready-made skill in one step
Hosted MCP connector works across Claude, ChatGPT, Claude Code, Cursor, Codex, OpenCode and Windsurf
Paid plans add connectors to Jira, Confluence, SharePoint and Drive with connected search
Version history and rollback for stored skills and memories
Sharing across teams and workspaces on paid plans
SSO, audit log export and a signed DPA on the Business plan for security-conscious organizations
How to use Rafter?
Sign up: Create a Rafter account and log in at heyrafter.xyz.
Add the connector: In your AI tool's settings, add Rafter as a custom MCP connector (or use the matching CLI or config entry for Claude Code, Cursor or Codex).
Sign in: Approve the sign-in window that opens the first time the tool calls Rafter.
Set the habit: Paste Rafter's instruction line into your tool's system or project instructions so it checks Rafter before answering and saves new context with add_memory.
Share a skill: Send a teammate's AI tool Rafter's setup URL to hand off a skill directly.
Add connectors on a paid plan: On Team or Business, connect Jira, Confluence, SharePoint or Drive for connected search.
Rafter's Use Cases
- Team onboarding
- Cross-tool skill sharing
- Connected document search
- Process documentation
