Loading
An MCP server lets an AI assistant such as Claude or ChatGPT use a service for you. Bolt Claw's is hosted at mcp.boltclaw.ai: add it, sign in, and your assistant can schedule posts, work your inbox and pull your leads with the permissions you chose. It asks you before anything that publishes, deletes or costs money. It opens with the launch.
You asked Claude:
Post the workshop reminder to every account on Friday.
Publishing needs your approval. Nothing has been posted yet.
Waiting for your yes
GET /v1/pages
Authorization: Bearer bc_live_...The same API the dashboard uses, with your own keys.
No key to copy and no file to edit. You sign in, you choose, and you stay in charge.
Add https://mcp.boltclaw.ai/mcp in your assistant's connector or MCP settings.
A Bolt Claw sign-in opens in your browser. For each area, choose whether the assistant gets no access, read, or read and write.
Ask it to schedule Friday's post, sum up today's comments or list this week's leads. Each tool is one call to the REST API with a key made for this connection.
Anything that publishes, replies, deletes or costs money stops and asks. Some actions wait for your approval in the dashboard, and an approval works once.
Every connection is listed under Settings, AI agents: rename it, change what it may do, or end it, and ending it there ends it.
An assistant never does more than the person who connected it could, and anything consequential waits for a yes.
A connection works with the access you gave it, clamped to your own. If your role changes, so does what the assistant can do, on its next call.
An assistant can never issue API keys, register webhooks, change your team or billing, or connect a social account. Ask, and it gives you the dashboard link.
An approval is bound to the exact request it was asked for, works once and lapses after a day. Paying a partner and disconnecting an account always ask.
What an assistant did, and when, is on a log of agent calls, readable in the dashboard and through the API.
The connection's key is made on the server and never passes through the conversation. Bolt Claw keeps only its hash.
Every tool is a call to the social media API, so scopes, rate limits and approvals apply exactly as they do to your own code.
The MCP server is on Free. It shares your plan's API rate, because every tool is an API call. Each person on your team can connect their own assistant.
Compare the plans$0, for good
$19 a month
$49 a month
$149 a month
It is built for any assistant that can add a remote MCP server, Claude and ChatGPT among them. You add the address, sign in to Bolt Claw in your browser, and choose what the assistant may do. Clients that take a bearer key can connect with an API key instead.
Not unless you let it. Tools that publish, reply, delete or cost money answer that nothing has been done until the assistant confirms, and publishing also waits for your approval in the dashboard by default. Your organisation can waive the switchable approvals; paying a partner, adjusting a commission, approving a partner and disconnecting an account always ask.
Issue API keys, register webhooks, change anybody's access, change billing or the plan, connect a social account, or approve its own requests. Those are refused whatever the connection was granted, and the assistant gives you the dashboard link instead.
Ninety days, renewed while it is in use, for up to a year. Signing out in your assistant, or ending the connection under Settings, AI agents, revokes its key.
Yes, on every plan. Its calls count against your plan's API rate, 20 requests a minute on Free, because every tool is one API call.
Yes. A read-only documentation server at boltclaw.ai/docs/mcp-server answers questions about the API without a key, so an assistant can look up a call before it makes one.
Connect Claude or ChatGPT on the free plan. Sign up with Google or an email address, and pay nothing until you choose a plan.