Skip to main content
Most agent-tool integrations require you to define every possible API call upfront or give the agent raw API access and lose visibility into what it does. The superglue CLI unlocks a third option: your agent discovers available systems, assembles the right integration call at runtime, and executes it. No pre-configuration of tools required.

​
How it works

Your agent uses the superglue CLI (sg) and the superglue skill. This skill provides a structured reference that teaches your agent all superglue CLI commands, auth patterns, tool schemas, and data flow conventions. At runtime, when your agent needs to interact with a system on behalf of a user:
  1. Discover available systems and their APIs
  2. Look up relevant endpoint documentation
  3. Run an inline tool config — a single command, no persistence
sg system list
sg system search-docs --system-id stripe -k "list invoices"

sg tool run --config '{
  "id": "get-invoices",
  "instruction": "Fetch recent Stripe invoices for a customer",
  "steps": [{
    "id": "list-invoices",
    "config": {
      "systemId": "stripe",
      "url": "https://api.stripe.com/v1/invoices?customer=<<customerId>>&limit=5",
      "method": "GET",
      "headers": {"Authorization": "Bearer <<stripe_api_key>>"}
    }
  }]
}' --payload '{"customerId": "cus_abc123"}'
No build, no save, no draft files. The tool config is assembled and executed in a single command. The agent returns the result to the user and moves on.

​
What this unlocks

Instead of pre-building a tool for every possible API call and data point your agent requests, the agent figures out the right call at runtime. A user can ask “get my last 5 invoices” or “update the billing email for customer X”, and the agent assembles the right tool config for each, using the same superglue Stripe system.

​
Full monitoring and observability

sg tool run creates a run record by default (skip with --no-create-run). You see:
  • What the agent built (full tool config)
  • What APIs it called and what came back (step results)
  • Whether it succeeded or failed
  • Duration and timing
Same run history dashboard, same monitoring, same alerting, irrespective of whether the tool was pre-built or run on the fly.

​
Secure credential access

The agent never sees raw secrets. It references credentials with <<systemId_credentialKey>> placeholders that are resolved server-side at execution time. API keys, OAuth tokens, database passwords — all stay in superglue. The agent only knows the placeholder names. OAuth tokens are automatically refreshed before each step execution. No token management in your agent code.

​
Permissioning and access control

The API key your agent uses determines what it can access:
  • Restricted API keys limit the agent to specific systems or tools
  • End-user system scopes control which systems each end user’s agent can reach
  • Environment isolation separates dev and prod credentials
These constraints are enforced at the system level regardless of what the agent tries to run.

​
Supercharging with context lookup

The real power comes when you instruct your agent to look up system configs and documentation before building inline tools. Instead of guessing endpoint shapes and field names, the agent queries the actual API docs stored in superglue: 
sg system find --id hubspot
sg system search-docs --system-id hubspot -k "create contact properties"
This returns relevant documentation chunks — endpoint URLs, required fields, auth patterns, response shapes. The agent uses this to assemble a correct tool config on the first attempt instead of hallucinating endpoints.

​
Setup

1

Install the CLI and Skill

Follow the CLI + Skills guide to install the CLI and skill for your agent framework.
2

Configure API access

Set environment variables for your agent’s runtime:
export SUPERGLUE_API_KEY="your-api-key"
export SUPERGLUE_API_ENDPOINT="https://api.superglue.cloud"
Limit what the AI agent can access via superglue role-based access patterns.
3

Register your systems

Set up the systems your agent should be able to reach — APIs, databases, file servers.
sg system create --name "Stripe" --template stripe \
  --credentials '{"api_key":"sk_live_xxx"}' \
  --docs-url "https://docs.stripe.com/api"
See Creating a system for more details.
4

Instruct your agent

Add detailed instruction on which systems to use, which tools to run and which CLI commands to invoke to your agent’s system prompts and Markdown reference files.

​
Next steps

CLI + Skills

CLI installation, skill setup, and project-specific agent configuration

CLI Commands

Full reference for sg tool run --config and all other commands

Permissioning

Restrict agent access with scoped API keys and end-user system scopes

Run History

Monitor and debug all tool executions, including inline runs