ThreadKeep_registerStart free with permission and securely stored keys.
A SAVE POINT FOR YOUR AI AGENT
Context ends. Work shouldn’t start over.
Give your agent a durable record of what it did, what it decided, and what comes next.
MCP + HTTP · Free to start · No model call per save
Listed in the official MCP Registry ↗
“Build passed. Staging is ready. Owner approval is still needed.”
ThreadKeep_checkpointGoals, decisions, completed work, blockers, next actions and file references.
version: 7 · savedRead the checkpoint, verify the current situation, and pick up the next action.
ThreadKeep_resumeBUILT FOR HOW AGENTS FAIL
Reuse the same request ID and payload after a timeout. You get the original save receipt.
Version checks reject a stale save. Read, reconcile, then continue.
Revocable keys can be limited to one task. Agent keys cannot buy anything or manage billing.
CONNECT ONCE. KEEP YOUR PROGRESS.
Help me connect ThreadKeep for this project. Read https://getthreadkeep.io/threadkeep/agents/guide.md. Prefer the OAuth MCP endpoint when your host supports it; otherwise use the API-key flow. Explain what data it stores and confirm permission and terms before enrolling. Start on Free, securely persist the credentials, and prove it works by saving and resuming a checkpoint. Ask before any paid upgrade; the Pro plan is $12/month.
In your agent’s MCP or integrations settings, add this remote server and choose OAuth. ThreadKeep opens a secure consent page. Create a free workspace or use your existing recovery key, save a backup, and approve the connection.
https://interesting-concepts-ventures.vercel.app/api/threadkeep-connectNo card required. The agent can save its first checkpoint immediately after authorization. Try resuming it in a new session to confirm it works.
Add a Streamable HTTP server in a host that supports custom headers. Agents can discover the tools without a key; saving and reading private state require one.
https://interesting-concepts-ventures.vercel.app/api/threadkeep-mcpX-ThreadKeep-Agent-Key to your agent key.With your owner’s authorization, use ThreadKeep_register over MCP or enroll through the HTTP API. Persist credentials outside your context window, configure the agent key and reconnect. No email or payment card needed.
Python quickstart · OpenAPI · MCP manifest · Example connection config
OAuth uses authorization code with PKCE, dynamic client registration, expiring access tokens and rotating refresh tokens. Your host controls tool installation and permissions. Compatibility depends on its supported connection features; this is not an app-directory listing or endorsement.
FROM FIRST SAVE TO EVERY NEW SESSION
ThreadKeep_registerStart free with permission and securely stored keys.
ThreadKeep_checkpointSave the current state with version and retry protection.
ThreadKeep_resumeRestore your task after a fresh session or handoff.
ThreadKeep_listFind the saved tasks your key can access.
ThreadKeep_usageSee actual remaining capacity and verify paid access.
ThreadKeep_plansCompare Free and Pro before choosing.
ThreadKeep_request_upgradePrepare an approval link. The user decides and pays.
START SMALL. PAY WHEN IT’S USEFUL.
For trying a real agent workflow.
For agents doing ongoing work.
No automatic upgrades or overages. Pro renews monthly until canceled by the owner. Both plans: 10 active keys, 60 requests/minute/workspace, 24 KB stored state per task. Beta capacity limits apply. Full limits. Conversation AI allowances are separate.
When more capacity would help, your agent can show its usage and offer a private Pro approval link. Review the workspace and $12/month price, confirm access to your owner recovery key, then pay through Stripe. Your agent checks that Pro is active and continues with the same key.
Decline and your saved tasks stay readable. No surprise upgrades, no per-save charges and no required subscription to another product.
Only the latest checkpoint is stored. Artifact links point to your files; they don’t back them up. A checkpoint records task state, not proof that an external action succeeded. Agents should verify external effects before retrying. Keep secrets out of checkpoints.
Save and read operations use Vercel and Supabase. They don’t send your task state to an AI model.