Connect any MCP server to your agents. No config files.
Paste a server URL, sign in once, pick the tools you want the agent to use. BetterClaw handles OAuth and puts the server’s tools next to your 99 connectors.
- Works with Streamable HTTP and SSE servers.
- OAuth sign-in handled in the dashboard.
- On every paid plan, from $19/mo.
What BetterClaw MCP is
MCP (Model Context Protocol) is the open standard that lets an AI agent call tools on an outside server: a database, a CRM, a docs search, your own internal API. BetterClaw is now an MCP client. Any server that speaks MCP over the web can be attached to a BetterClaw agent, and its tools show up in the agent’s toolbox alongside Gmail, Slack, Telegram and the rest.
That means you are not limited to the integrations we built. If a vendor ships an MCP server, or your team writes one, your agent can use it the same afternoon.
How it works
Add the server.
Open an agent, go to Tools, choose "Add MCP server" and paste the server’s URL. BetterClaw detects whether it uses Streamable HTTP or SSE.
Sign in once.
If the server needs OAuth, BetterClaw opens the sign-in in your browser and stores the token encrypted with AES-256, auto-purged from agent memory after 5 minutes like every other secret. If it needs an API key or a header instead, paste it in the secrets field; it is never shown again.
Pick the tools.
BetterClaw lists every tool the server exposes. Tick the ones this agent should have and save. The agent can use them in the next message, and you can change the selection later without reconnecting.
No YAML, no JSON5, no restart, no callback URL to expose. Paste the URL, sign in, tick the tools, and the agent has them on its next message.
If you are coming from OpenClaw or Hermes
Both OpenClaw and Hermes Agent support MCP, and both do it through config files on the machine running the agent. The outcome is the same as on BetterClaw; the work to get there is not. This is what the setup looks like on each, as of October 2026.
Where the server is defined
A form in the dashboard: URL, auth, tool checkboxes
mcp.servers block in the OpenClaw config (url, transport, command, args, cwd, timeouts, toolFilter)
mcp_servers block in ~/.hermes/config.yaml (command, args, env, cwd, url, headers, client_cert, client_key)
Transports
Streamable HTTP and SSE (stdio not supported yet)
Streamable HTTP, SSE, stdio
stdio and HTTP (with mTLS)
OAuth
Sign in from the dashboard; BetterClaw handles the callback and token refresh, nothing on your machine
Sign in from Settings > MCP, or openclaw mcp login; browser callback only works from the gateway’s own HTTPS, loopback or Tailscale address
auth: oauth with PKCE and refresh via the Python SDK; prints an authorize URL and waits on a local callback; headless hosts paste the redirect back or use an SSH port forward
Choosing tools
Checkboxes per tool, changeable later without reconnecting
toolFilter include/exclude patterns in config
tools.include / tools.exclude in config (include wins if both set)
Applying changes
Save; live on the next message
openclaw mcp reload; a gateway running elsewhere needs its own reload, config publish or restart
Auto-reload in a running session with a 30-second timeout (too short for OAuth, so log in from a second terminal), or /reload-mcp
Checking it works
Enable a tool, assign it to an agent and try it in chat
openclaw mcp doctor --probe, openclaw mcp status --verbose
hermes mcp test <name>; servers that failed to connect are named in agent.log
Sources for the OpenClaw and Hermes columns: docs.openclaw.ai/tools/mcp and the Hermes Agent MCP user guide, both read on 12 Oct 2026. Comparing in more depth? BetterClaw vs OpenClaw and BetterClaw vs Hermes.
What you can connect
Anything that publishes an MCP endpoint. Remote servers published by vendors (the kind listed as “remote MCP server URL” in their docs) work as-is.
MCP tools and BetterClaw connectors can be used in the same agent. A support-triage agent can read the ticket from Gmail, look up the customer in your CRM over MCP, and reply in Slack, in one run.
MCP and local models
If you run your agents on Ollama, LM Studio or llama.cpp through BetterClaw Desktop, MCP servers still work with any model that supports tool calling: the model runs on your machine, the MCP server runs wherever it runs, and BetterClaw brokers the tool calls between them. Tool-calling quality depends on the model; the Desktop app shows which local models support tool calls.
Get BetterClaw DesktopSecurity
MCP tools go through the same trust controls as every other BetterClaw tool.
Each agent sees only the servers attached to it.
Tokens and headers are stored encrypted with AES-256, auto-purged from agent memory after 5 minutes, and never shown again after saving.
You can detach a server at any time; the agent loses the tools immediately.
BetterClaw is ISO 27001 certified and publishes its subprocessors.
What we do not do
We do not run arbitrary commands from a server’s manifest on your behalf.
We do not auto-install catalogue entries.
You add servers explicitly.
Plans
MCP is available on every paid BetterClaw plan, from Basic at $19 a month, and on AppSumo lifetime Tier 3 and above. MCP tool calls use credits the same way built-in tools do. There is no cap on how many servers you add, but every enabled tool adds to the agent’s input tokens, so enable only the ones it needs.
See pricingMCP is included on AppSumo lifetime Tier 3 and above. See the AppSumo deal
Paid plans start at $19 a month.
BetterClaw MCP: frequently asked questions
Does BetterClaw support MCP?
Yes. BetterClaw is an MCP client: you can attach any MCP server that is reachable over the web (Streamable HTTP or SSE) to any agent, and its tools appear next to your other connectors.
How do I add a custom MCP server?
Open the agent, go to Tools, choose Add MCP server, paste the URL, complete sign-in if the server needs it, tick the tools you want, and save. No config file and no restart.
Which transports are supported?
Streamable HTTP and SSE. Stdio servers, which run as a local process, are not supported yet.
Does it handle OAuth?
Yes. Servers that use OAuth sign in from the dashboard; BetterClaw handles the callback and token refresh, so nothing runs on your machine.
Can I choose which tools the agent sees?
Yes. Every tool the server exposes is listed with a checkbox, and you can change the selection later without reconnecting.
Is MCP included in the AppSumo lifetime deal?
Yes, on Tier 3 and above.
Is there a limit on how many MCP servers I can add?
No. Keep in mind that every tool you enable is described to the model on each message, so more servers and tools mean more input tokens per run. Enable only the tools an agent needs.
Do MCP tool calls use credits?
Yes, the same way built-in tools do.
How is this different from OpenClaw’s MCP support?
The capability is the same: both connect to MCP servers. OpenClaw configures them in its config file and manages OAuth, reloads and health through the openclaw mcp CLI. BetterClaw does the same in a form in the dashboard, with the OAuth callback handled for you. The comparison table above has the details.
What is the difference between an MCP server and a BetterClaw connector?
A connector is an integration we built and maintain (Gmail, Slack, Telegram and 96 more). An MCP server is an integration anyone can publish using the open standard. Agents can use both in the same run.
Can BetterClaw be used as an MCP server from Claude or Cursor?
Not today. BetterClaw is an MCP client. This is planned.
Attach your first MCP server in under five minutes.
Start with Basic, $19/mo7-day money-back guarantee · Cancel anytime