toolspoke
Start free

Connect AI agents to your team's tools, safely.

Toolspoke sits between Claude, Cursor, Codex and the tools they use. You choose what each agent can do, and every call is logged.

One address for every client.

Start freeView docs

102 integrations · 1,143 actions

102 integrations, ready to install

Each one was set up, credential-mapped and action-classified by hand. Point the toolkit builder at anything missing.

25 of 102 integrations

Browse all 102 integrations

Choose what each agent can do

A project decides which of a toolkit's actions it allows. An agent key gets that list, or a slice of it. Blocked actions are filtered out before the tool list reaches the client.

Gmail in a Toolspoke project: ten actions grouped read, write and delete, each with a switch, headed 8 of 10 on.
Swipe sideways.

Agents never see your API keys

Credentials are stored on the project and attached inside the gateway as a call goes out. The agent receives the result, and secrets are masked in the record.

A Toolspoke audit record for an aws invoke call, its api_key argument shown as [REDACTED] beside the response.
Swipe sideways.

Every call is logged, request and response included

Open one and you get what the agent sent, what the tool returned, the rule that allowed it and the credential that ran it.

LiveActivity: every call, allowed and refusedSelect a call to hold the stream

github · list_pull_requests

Claude Desktop acting as Ashiq in Engineering

200 ok183ms
{
  "owner": "toolspoke",
  "repo": "toolspoke",
  "state": "open"
}

The arguments the agent sent, whole. Anything that looks like a secret is masked on read-back.

The Toolspoke activity feed: fourteen tool calls with agent, user, project, status and duration, one of them failed.
Swipe it sideways.

Add any API from its docs

Point the toolkit builder at a documentation URL, an OpenAPI file, a remote MCP server or a CLI. Every method, path and tool name is copied from a deterministic extractor rather than written by the model.

The Toolspoke toolkit builder, with a documentation URL typed into its source composer.

Set up in five minutes

  1. Create a workspace

    No card. Free starts with 5,000 credits, topped back up every month.

  2. Install toolkits into a project

    A project holds the toolkits and their credentials. Set which actions it allows.

  3. Point your client at the gateway

    Paste the config, or sign in from Claude, Cursor, VS Code, OpenAI and OpenCode.

The agent is handed five tools rather than all 1,143 actions. It calls search_tools to find the rest.

mcp.json

{
  "mcpServers": {
    "toolspoke": {
      "type": "http",
      "url": "https://toolspoke.com/mcp"
    }
  }
}
Start freeConnect your agent

Free to start, $19 per seat for teams

Every plan gets the whole gateway. The plan decides how many credits arrive with each seat, and one credit covers one attempt at a tool.

Compare plans and credit top-ups

In every plan

  • Every marketplace and custom tool
  • Per-user policies on every action

Free

Run real agents on the complete gateway, free every month.

$0

forever

  • 5,000 credits every month
Start free

Pro

Most teams

For teams running agents in production.

$19

per seat, per month

  • 50,000 credits per seat each month
Choose Pro

Enterprise

Managed hosting, SSO, extended retention and an SLA.

Custom

annual agreement

  • Credit volume to match your usage
Talk to us

Start with the free plan

2 seats, 5,000 credits a month, no card.

Start free