CoreSpeed FAQ: plans, credits, clients, security, what it does not do
Short answers on what CoreSpeed is, which agents work, what is free, where tokens live, who approves, whether it runs agents, and how to leave.
CoreSpeed is one remote MCP server that gives every agent you run the tools of the apps your organization connected, durable memory, media generation, live web search and public social data. You sign in once. This page answers the questions people ask before and right after that first sign-in, in the order they tend to ask them.
What it is
What is CoreSpeed?
One MCP server at https://api.corespeed.io/mcp, over Streamable HTTP. Sign in through the browser or send an API key. Every agent you run then gets the same tools, the same memory, the organization's spend caps and one activity trail. The organization is the unit of ownership: connections, memory, budgets, policy and the audit trail live there, not in any one agent. The Introducing CoreSpeed post gives the longer version.
Does CoreSpeed run my agents?
No. CoreSpeed equips the agent you already run: Claude Code on your laptop, Codex in your terminal, a nightly job you operate on your own server. The agent signs in to CoreSpeed; CoreSpeed does not start, host or schedule it.
Which agents and clients work with it?
Claude Code, Codex, Cursor, Copilot in VS Code, OpenClaw, Hermes, claude.ai and Claude Desktop (as a custom connector on Pro, Max, Team and Enterprise plans), ChatGPT (as a custom connector under Developer mode, on Plus and Pro on the web), and any client that reads an mcpServers map. Per-client steps are on the MCP server page.
How do I set it up?
Give your agent this line in chat:
set up https://corespeed.io/SKILL.mdThe agent fetches the skill, adds the server at user scope, signs you in, keeps the skill and asks which apps to connect. The manual form for Claude Code is claude mcp add corespeed https://api.corespeed.io/mcp --transport http --scope user, then /mcp to authenticate.
Plans and credits
What are credits, and what is free?
Every metered action is priced in credits, and per-action prices are on the pricing page. Memory operations and discovery (tools/list) cost 0 credits. When your first organization is created you receive 3,000 free credits, no card required, once per user.
What does each plan include?
| Plan | Price | Members | Memory | History | Credits |
|---|---|---|---|---|---|
| Free | $0 | 1 | 200 MB | 30 days | 3,000 free credits at signup |
| CoreSpeed Pro | $20 per month | up to 25 | 2 GB | 1 year | 10,000 each month, plus top-ups |
| Enterprise | custom | Book a call |
Do credits expire?
Signup credits expire 90 days after the grant. Pro plan credits reset monthly and are spent before the top-up balance. Top-ups of $5 to $1,000 never expire. The Free plan has no top-up.
What happens when I run out?
Metered calls are refused before the tool runs, as HTTP 200 with isError: true and the code payment_required. Discovery, account reads and key management keep working. An org admin adds credit and later calls run. A monthly cap on one API key works the same way for that key, with key_spend_limit_exceeded. Details are on the billing page.
Security
Where do my app tokens live?
In CoreSpeed's custody. When you connect an app, CoreSpeed stores the credential, refreshes it when the provider allows, and exposes namespaced tools such as slack__post_message. The agent receives tools, never the token. Disconnecting deletes CoreSpeed's copy.
Who can see what my agent did?
You see your own actions in Dashboard → Activity; org admins see the organization. A record holds the action, the actor, the outcome and, for failures, the code and the text the agent saw. Prompts are not audit payloads, and sensitive request content is not copied.
Who approves an agent's action?
If your organization turned Smart Approval on, the member who made the call or the owner of the agent or API key that made it. An agent cannot approve its own action. Admins see every card, and seeing is not deciding. Cards arrive in the dashboard queue, by email, or in Slack.
Does every action go through approval?
No. Reads are never gated. Writes are judged against a policy you wrote in plain English; the actions the policy names pause, and on the default Balanced setting an ordinary write the policy never mentions runs. The gate sees only calls made through CoreSpeed, and a new organization has no policy and no gate until an admin turns it on.
Which credential should an unattended job use?
An sk-csa- agent key. It belongs to an agent identity with an owner, reaches the organization's shared connections only, and can carry a monthly spend cap. A member's sk-cs- key acts as that member and reaches their private connections too.
Data and leaving
How do I delete memory?
memory__delete_memory removes one memory and memory__clear_all_memory removes all of yours. Where memory maintenance has rolled out to your organization, memory__archive_memory takes a memory out of search without deleting it and memory__restore_memory brings it back.
How do I leave?
Disconnect each app in Dashboard → Connectors, which deletes CoreSpeed's copy of the credential. Revoke your API keys and retire any agents. Remove the corespeed entry from your clients. Memory you want gone is yours to delete first.
Where do I get help?
The docs at https://corespeed.io/docs, the machine-readable index at https://corespeed.io/llms.txt, and the Discord at https://discord.com/invite/NdBHBvDRXX.