clawband
native MCP server · 128 providers

Set your agents loose. Keep them accountable.

Pick the tools your agent can call. Pin each parameter to a constant. Hand it a key that can only post to #dev-log — not refund Stripe charges, not read your inbox. Across Stripe, GitHub, Gmail, Notion, Linear and 123 more, with a full audit log.

Start free

Free tier. Usage-based pricing after that. No card required.

terminal
# Add Clawband to Claude Code (or any MCP client)
claude mcp add clawband \
  --transport http \
  --url https://api.clawband.io/mcp \
  --header "Authorization: Bearer cb_live_a8f3…"

# Your agent now sees only the tools you ticked,
# with the parameters you pinned. Nothing else exists.

The problem

Provider APIs let your agent do everything you can. Including the things you'd never tell it to.

Raw API keys give your agent every endpoint, every value

Drop your Stripe key into an agent's config and it can refund any charge, for any amount, to any account. Provider APIs don't let you say 'only refund up to $50, only from this customer.' Your agent inherits everything you can do, at any value it picks.

Every service wants its own OAuth dance

Wiring an agent to Gmail, Notion, GitHub, Linear, and Stripe means five OAuth apps, five sets of scopes, and five token refreshes, all before you've written a single prompt. And the moment you swap agents, you do it all again.

No idea what the agent actually did

When the agent sends the wrong email, closes the wrong issue, or refunds the wrong charge, you're left grepping through twelve provider dashboards. There's no single per-call log of what it tried, what it sent, and what came back.

How it works

Connect once. Grant narrowly. Audit everything.

  1. 01

    Connect your accounts

    Sign into Stripe, Gmail, GitHub, Notion, or whatever else you use, once, in the Clawband dashboard. Tokens are encrypted and stay in the gateway. Your agent never sees them.

    stripe   · connected
gmail    · connected
github   · connected
notion   · connected
  2. 02

    Pick the tools. Pin the parameters.

    Issue a Clawband key per agent. Tick the exact tools it can call. Lock parameters to constants — only this Slack channel, only this Notion database, max refund $50 — so a prompt injection can't widen the surface on a bad day.

    agent: my-coding-agent
  + github.create_pr
  + linear.update_issue
  + slack.post_message
      channel = #dev-log   # pinned
  3. 03

    Point your agent at one endpoint

    Drop the key into Claude Code, OpenClaw, Cursor, or your own runtime via MCP, or POST JSON to /v1/invoke. Clawband enforces the policy and writes a per-call audit log you can review.

    POST /mcp
{
  "method": "tools/call",
  "params": { "name": "github.create_pr", ... }
}

Code

One line to wire up. The tools your agent sees are the tools you ticked.

Clawband speaks native MCP and a plain HTTP endpoint. The tools/list your agent gets back contains only what you ticked, with the parameter schemas you pinned baked in. Anything else is invisible — and uncallable.

terminal
# MCP setup (one line)
claude mcp add clawband \
  --transport http \
  --url https://api.clawband.io/mcp \
  --header "Authorization: Bearer cb_live_a8f3…"

Integrations

128 services your agent can use.

See all 128 integrations
Stripe
Slack
GitHub
Notion
Linear
HubSpot
Shopify
Google Drive
Gmail
Google Calendar
Zoom
Microsoft Teams
Discord
Twilio
Asana
Jira
Trello
ClickUp
Mailchimp
SendGrid
Vercel
Cloudflare
AWS
Datadog

+ 104 more, across AI, calendar, email, cloud storage, dev tools, finance, smart home, and more.

Security & policy

If your agent's key leaks, the blast radius stays small.

Policy at the edge

Per-agent tool allowlists

Each Clawband key carries an explicit list of tools it can call. Everything else is invisible to that agent, even if the provider's API supports it.

my-coding-agent:
  - github.create_pr
  - linear.update_issue

Parameter pinning

Lock specific parameters to constants. The agent can't override them, even if it tries, even if a prompt tells it to.

slack.post_message:
  channel: "#dev-log"   # pinned

Minimal OAuth scopes

When you connect a provider, Clawband requests only the OAuth scopes the tools you've enabled actually need. Nothing speculative.

github scopes:
  repo:status, pull_request

Observability & safety

Response scrubbing

Some provider APIs return credential material (webhook signing secrets, API keys, embed tokens) inside otherwise routine responses. The gateway strips that material before the agent sees it, so a misbehaving agent can't read its way around the policy.

// upstream response
{ "id": "whsec_...", "secret": "whsec_8a…" }

// what the agent receives
{ "id": "whsec_...", "secret": "[redacted]" }

Full audit log

Every call: timestamp, which agent key, which tool, the exact parameters sent upstream after policy was applied, upstream status, and a hash of the response. Browse it in the dashboard or stream it to your own sink.

2026-05-26T11:04:12Z  my-coding-agent
  slack.post_message  channel=#dev-log
  upstream=200  hash=sha256:7c…

FAQ

Common questions.