CoreSpeed - One MCP for everything your agents need: apps, memory, tools
by•
CoreSpeed is an operating system for agents. Connect apps once and bring your accounts and shared or private memory to Claude Code, Codex, Cursor and other MCP agents. Use multiple accounts per app, including X, with web search, social research and media generation built in. CoreSpeed holds app credentials. Budgets, activity logs and Smart Approval (beta) keep agents in check. Planned: Agent Drive, Mail, Pay, sandboxes, and more. Everything your agents need, through one MCP endpoint.
Replies
Hey Product Hunt! 👋 We're the founding team of CoreSpeed.
We kept hitting the same wall: our agents were capable, but getting them to do real work meant wiring up integrations, managing credentials, and rebuilding context. After rebuilding that foundation across three products, we realized the foundation itself was the product.
So we're building CoreSpeed, an operating system for agents. Set it up once, and every MCP agent you sign in, whether Claude Code, Codex, Cursor or your own, works from the same setup:
- Your connected apps, with multiple accounts per app, like personal and company accounts on X
- Saved memory, shared with your team or kept private
- Built-in web search, social research, and image, audio and video generation. Think of this layer as OpenRouter for agent tools: no separate API keys or provider subscriptions. Just pay as you go through CoreSpeed.
- App credentials held by CoreSpeed, never handed to your agents
- Budgets, activity logs and Smart Approval (beta), where you write in plain English which actions need your OK first
Planned, not live yet: Agent Drive, Mail, Pay, sandboxes, and more.
Everything your agents need, through one MCP endpoint. Give this to your agent to start for free:
set up https://corespeed.io/SKILL.md
@miltonheyan How does smart Approval decide which actions need human confirmation ?
@dipanshu_kushwaha5 Hi Dipanshu, thanks for comment! Smart Approval checks write actions against rules you set in plain language. If approval is required—or it’s unsure how a rule applies—it pauses and asks you rather than guessing. (Jev evaluates the tool call and your rules, it uses a confidence score. If confidence score falls below the configured threshold, it asks for human confirmation instead of guessing).
More on how it works: https://corespeed.io/blog/introducing-smart-approval
@miltonheyan @dipanshu_kushwaha5 Hi Dipanshu, for example, if you are dealing with a customer complaint involves a small refund, your agents handle the ticket and make a refund without human in the loop when it stays in the "smart approval" zone, when a larger amount gated by smart approval rules.
Checkout our Smart Approval video here: https://youtu.be/-MOXwSdUSbE?si=PB-PCLYS1KhLjZND
@miltonheyan Congrats on the launch! The core idea is easy to understand: connect apps once, then make them available to different coding agents.
@miltonheyan @vipul_kumar1280 Thanks Vipul! Trying a new agent shouldn't mean reconnecting all your apps and starting its memory from scratch. That's what we're trying to fix with CoreSpeed.
@miltonheyan Can I inspect the exact action before approving it? I'd want to check both the account and the destination.
@miltonheyan @amrita_prajapati2 Yep Amrita, you can check the tool call and its arguments in the dashboard before approving it. Smart Approval cards are still pretty basic today—we're working on making those details easier to see right in the card.
@miltonheyan A tool is at its best when the engineers who built it use it every single day. Here's a 2-minute look at the kind of day we built CoreSpeed for: one support ticket, six apps, and a switch from Claude Code to Codex halfway through, with nothing to re-explain. https://youtu.be/WkJX4gyd_Yk
@scoootscooob Thanks for your support Scott❤️
@scoootscooob thanks scott
The activity log seems like a useful place to start a demo. It could show what the agent accessed, changed, and spent.
@wys1010 Thanks Yansong, will try it
@wys1010 I like that angle. Seeing the actual tool calls and costs makes a demo much more concrete than just watching an agent say “done.”
How does an agent choose among the available media-generation tools? Can the user specify a preferred provider?
@eexlkuang_se Yep, you can ask for a specific model. The agent can check the available image, video and audio models and their inputs before choosing, so you don’t have to leave that choice entirely up to it.
keeping the app credentials away from agents makes sense. if a smart approval rule is ambiguous, does the action wait for me or go through?
@jeetendra_kumar2 Thanks Jeetendra! If the evaluator isn’t confident how your rule applies, the action pauses for your approval rather than going through. You can approve or reject it via Slack, email, or the dashboard.
@miltonheyan ah nice, pausing for approval instead of blocking outright is the right call - does the approval ping show what rule it tripped on?
can i ask u one question? how is it different from something like Composio?
Wait this actually solves one of my biggest annoyances with coding agents lol. Every time I try a new one I end up reconnecting and configuring everything all over again. The “set it up once” idea makes so much sense.
@contact_paradigm Thanks cutie Clover, we may have more cooperation soon, use Paradigm for learning, use Claude Code/CodeX for coding, and use Muse for daily life, CoreSpeed brings everything across the Agents and you never set up everything again and again!
@axelkane The activity trail focuses on the action and outcome, with sensitive metadata redacted before storage. It isn’t intended to keep a copy of your prompts. I still need to confirm the custom field-exclusion part before giving you a definite yes on that.
We keep raw tool arguments and conversation prompts out of the activity log. It records who called which tool, when, the outcome, and any cost. Failed calls also include an error code and a brief error message, so you can troubleshoot without storing the full request payload.
Connecting an app once and reusing the same accounts and memory across Claude Code, Codex and Cursor solves a real annoyance, since everyone ends up wiring the same integrations over and over. Supporting multiple accounts per app is a thoughtful detail. How do you keep shared versus private memory separated between agents?
@karimbenkeroum Good question Ben, for the memory part, the default option is "privet", if you want to switch it to "shared", you can just tell your agent"make this memory to be shared", or you can visit the CoreSpeed dashboard to switch it to "Org Shared".
@karimbenkeroum One distinction worth adding: private memory belongs to your identity, not to a particular agent client. So Claude Code and Codex signed in as you can use the same private memories, while another teammate cannot. Shared memories are currently available across your organization. We’re also rolling out more granular controls so you can choose who to share them with.
the "credentials held by CoreSpeed, never handed to your agents" line is the part that actually matters to me, that's the difference between a tool and a liability. we've been wrestling with a similar split at Dial on the data side - wanting the agent to act on behalf of the business without the agent (or us) ever seeing the raw credential. question on Smart Approval: when an agent hits an action that needs OK mid-task, does it block and wait, or queue the rest of the plan and come back to that one step once you approve
@galdayan That’s up to the harness, Gal. CoreSpeed holds the specific tool call for approval. Your harness can wait on it or keep working on independent steps—the rest of the plan isn’t blocked by CoreSpeed.
@spencer_zhao that's the right call actually, better than CoreSpeed guessing a default. a half-finished plan sitting on pause because of one approval is its own kind of silent failure if the human doesn't check back for a while - is there a nudge on the harness side (webhook, notification) when an approval's been sitting for a long time, or is it purely poll-and-wait until someone looks at the dashboard
@galdayan We designed it so requests don’t hang indefinitely: if the agent stops waiting or the approval times out, the request expires. The agent can give up or make a fresh attempt, which triggers a new Slack or email notification if enabled. So the nudge comes with a fresh approval request, rather than repeated reminders for an old one.