grphly

grphly

Push in Cursor.
Fetch in Claude or ChatGPT.

One knowledge base — not five conflicting mental models. Push chat summaries, architecture decisions, and research dumps from Cursor, Claude, or ChatGPT. Your teammates fetch the same context in whatever AI they use.

Connect your AI — free beta See real examples

One memory layer across the tools you already pay for

The cost of forgetting

You’re paying for AI
that has amnesia.

Context windows reset. New chats wipe the slate. Docs rot in Notion. Slack threads die. So you paste the same architecture note — again — and burn tokens explaining what you already decided.

Without grphly

“We use Postgres, right?” — asked for the 4th time this week. Junior joins, spends half a day reconstructing decisions from old PRs. You switch from Cursor to Claude and start over.

With grphly

Push the decision once. Any teammate, any assistant, any day — kb fetch returns the same source of truth. Less re-prompting. Fewer wrong assumptions. Faster shipping.

Why teams switch

Knowledge that survives
the chat.

  1. Save hours every week Stop copy-pasting “project context” into every new thread. Your AI already knows the stack, constraints, and “why we did it this way.”
  2. Keep assistants in sync Push in Cursor. Fetch in Claude Code or ChatGPT. One knowledge base — not five conflicting mental models.
  3. Onboard without a lecture Invite a teammate. They sign in with Google, connect grphly, and ask the project what matters. No 45-minute voice call.
  4. Feel like git — not another wiki Type kb push / kb fetch in chat. No new app to live in. No ceremony. Just durable memory where you already work.

Examples

Push whole sessions.
Not just one-liners.

kb push takes any durable text — a sentence, a design dump, or a summarized chat from another LLM. Fetch later with a real question.

Cursor session → Claude Code

After a long Cursor thread, push the summary into grphly. Next day in Claude Code, fetch it — same stack decisions, no re-brief. 

kb push """
Session summary (Cursor, auth refactor):
- Auth0 + Google social only; no local passwords
- Access tokens audience = https://grphly-dev.miless.app/mcp
- Callbacks: localhost:8787 + Claude auth_callback
- Rejected: rolling our own JWT store
Open: invite email copy still draft
"""

kb fetch What did we decide about auth and callbacks?

ChatGPT research → team Cursor

Cofounder researches pricing and competitors in ChatGPT, pushes the condensed notes. You open Cursor and fetch the market context without reading their chat log. 

kb push """
ChatGPT research dump — competitor memory tools:
- Mem0: strong SDK, weaker shared-team KB story
- Our wedge: shared kb_id + invite by Google email
- Pricing hypothesis: free solo, paid seats for orgs
Sources discussed: mem0.ai, Reddit MCP threads
"""

kb fetch What's our wedge vs Mem0 and pricing hypothesis?

Claude design review → agency handoff

Client constraints come out of a Claude review. Push the full brief into a client KB. Anyone on the account fetches rules instead of reinventing the stack. 

kb create acme-rebrand "Acme Q2 site"

kb push """
Claude design review — Acme Q2:
Must: custom CSS only, no Tailwind, WCAG AA
Deploy: Vercel, preview URLs for stakeholders
Tone: quiet, technical, no emoji in UI copy
Out of scope: mobile app, CMS migration
"""

kb fetch What CSS and deploy constraints does Acme have?

Pricing

14 days free.
Then one simple plan.

Start with a full trial from your first Google sign-in. When it ends, subscribe once — unlock every kb_* tool again.

Pro

$19 / month

  • 14-day free trial on first login
  • Unlimited knowledge bases
  • Push & fetch across Cursor, Claude, ChatGPT
  • Invite teammates by Google email
  • One seat = one Google account
Start free in your AI

After trial: in chat run kb upgrade — grphly returns a Stripe Checkout link. Payment unlocks your account via webhook (no need to reconnect OAuth).

Commands

If you can type git,
you can use grphly.

No dashboard homework. In chat, say what you mean — the model calls the MCP tool. Push knowledge like you push code. 

kb list
kb create <id> [name]
kb use <id>
kb push <text>
kb fetch <question>
kb invite <email> [read|write]
kb members
kb revoke <email>
kb delete <id>
kb pushstore summaries, decisions, research dumps
kb fetchask across what any LLM pushed
kb invitegive a teammate the same brain
kb listsee every knowledge base you own
kb createone KB per product, client, or idea
kb membersknow who can read or write

Get value in minutes

Connect once.
Never start from zero again.

  1. Connect your AI Add grphly as an MCP server. Sign in with Google. Under two minutes on Cursor.
  2. Create a knowledge base kb create my-product — one place for that project’s durable truth.
  3. Push what matters Architecture notes, chat summaries, “never do X.” Anything durable you don’t want trapped in one model’s context window.
  4. Invite the people who ship with you They connect the same MCP, join the KB, and stop pinging you for context.

Setup

Connect your AI.
Start pushing today.

Same endpoint everywhere. Pick your client, paste the snippet (includes the OAuth client id), sign in with Google — then run kb push on the first fact you’re tired of repeating.

OAuth client id (Auth0 native / MCP clients): 0duBflkrH7BfwctBStAM8h6AAzAtbexD

  1. Open Cursor Settings → MCP, or edit .cursor/mcp.json.
  2. Paste the snippet below (URL + CLIENT_ID) and save.
  3. Refresh the server, then sign in with Google when prompted.
  4. In chat: kb create my-project then kb push …
.cursor/mcp.json 
{
  "mcpServers": {
    "grphly": {
      "type": "http",
      "url": "https://grphly-dev.miless.app/mcp",
      "auth": {
        "CLIENT_ID": "0duBflkrH7BfwctBStAM8h6AAzAtbexD"
      }
    }
  }
}
  1. In the Agent panel, open … → MCP Servers → Manage MCP Servers.
  2. Choose View raw config to open mcp_config.json.
  3. Add the snippet (uses serverUrl). Complete Google OAuth when prompted.
  4. If auth asks for a client id, use the OAuth client id above.
  5. Refresh MCP, then kb list.
~/.gemini/antigravity/mcp_config.json 
{
  "mcpServers": {
    "grphly": {
      "serverUrl": "https://grphly-dev.miless.app/mcp"
    }
  }
}
  1. Run the command below (includes --client-id for Auth0).
  2. Inside Claude Code, run /mcp and complete Google sign-in.
  3. Confirm grphly tools appear, then try kb list.
terminal 
claude mcp add --transport http \
  --client-id 0duBflkrH7BfwctBStAM8h6AAzAtbexD \
  grphly https://grphly-dev.miless.app/mcp
  1. Open Claude → Settings → Connectors (or MCP servers).
  2. Add a custom connector with the MCP URL below.
  3. If Claude asks for an OAuth client id, paste the client id from above.
  4. Authorize with Google, then type kb list.
MCP server URL 
https://grphly-dev.miless.app/mcp
OAuth client id 
0duBflkrH7BfwctBStAM8h6AAzAtbexD
  1. ChatGPT Settings → Connectors → enable Developer Mode (where available).
  2. Add a custom connector named grphly. Paste the MCP URL.
  3. Choose OAuth. If asked for a client id, use the one above.
  4. Sign in with Google, enable the connector, try kb list.

Custom connectors may require a paid ChatGPT plan with Developer Mode. Auth0 must allow ChatGPT’s OAuth callback URL for that client.

MCP server URL 
https://grphly-dev.miless.app/mcp
OAuth client id 
0duBflkrH7BfwctBStAM8h6AAzAtbexD

Your next chat
doesn’t have to be blank.

Connect grphly once. Push the facts you’re tired of repeating. Tomorrow’s AI — and your teammates — already know.

Connect your AI — free beta