Tools & Integrations

Tools are the capabilities an agent can invoke during a task. Without tools, an agent can only reason and generate text. With tools, it can take action: search the web, read files, call APIs, send messages, and more.

You control exactly which tools each agent has access to. Only enable what the agent actually needs.

Configuring tool access

Tool access is configured in the tools.allowed array in the agent config:

"tools": {
  "allowed": [
    "web_search",
    "memory_search",
    "memory_store",
    "send_message"
  ]
}

Any tool not listed here is unavailable to the agent, even if the platform supports it.

Built-in tools

`web_search`

Searches the web and retrieves current information. The agent formulates search queries autonomously and synthesizes results into its response.

Use cases: News monitoring, competitor research, fact-checking, current events

No additional configuration required.

`memory_search`

Queries the agent’s long-term memory store using hybrid semantic + keyword search. The agent calls this automatically at the start of tasks when memory search is enabled.

Use cases: Recalling past decisions, referencing prior research, continuity across sessions

"memory": {
  "search": {
    "enabled": true,
    "topK": 5
  }
}

`memory_store`

Saves information to the agent’s long-term memory store. The agent calls this when it determines something is worth remembering for future sessions.

Use cases: Storing client preferences, saving research findings, building a knowledge base over time

This tool is paired with memory_search — enable both together.

`send_message`

Sends a message via a connected notification channel (Telegram, Slack).

Use cases: Proactive alerts, scheduled report delivery, escalating to a human

The agent can call this during any task — not just cron jobs. If it detects something urgent, it can send an alert unprompted.

Prerequisite: At least one notification channel must be connected under Settings → Notifications.

`read_file`

Reads files from the agent’s workspace directory. Useful for agents that operate on data files, templates, or configuration documents you upload.

Use cases: Processing CSV reports, reading templates before drafting, loading context documents

agent-workspace/
├── SOUL.md
├── data/
│   └── monthly-report.csv   ← agent can read this
└── templates/
    └── email-template.md    ← agent can read this

`write_file`

Creates or updates files in the agent’s workspace directory.

Use cases: Drafting documents, generating reports, maintaining a running log

`run_script`

Executes scripts in the agent’s container environment. The agent can write and run scripts to process data, transform files, or perform computations.

Use cases: Data processing, format conversion, complex calculations, automation scripting

`http_request`

Makes HTTP requests to external APIs. The agent can call REST APIs, retrieve data, or trigger external systems.

Use cases: Pulling data from CRMs, triggering webhooks, posting to external services, reading from internal APIs

// Example: agent calls your internal reporting API
// Agent invokes: GET https://api.yourapp.com/reports/weekly

Credential management: Never put API keys in the SOUL file. Use environment variables configured in the agent’s deployment settings, which the agent can reference by name.

n8n workflow integration

AgenFleet integrates with n8n for deterministic automation that doesn’t need AI judgment.

Pattern: Agent handles the reasoning → triggers an n8n workflow for the execution

Example:

Research agent identifies a lead worth following up on
Agent calls http_request to hit your n8n webhook endpoint
n8n handles the CRM update, email send, and task creation — deterministically

This keeps AI where it adds value (judgment, synthesis, interpretation) and deterministic automation where it’s more reliable (data entry, notifications, system updates).

MCP (Model Context Protocol) connections

AgenFleet supports MCP servers for extending agents with custom tool sets. MCP allows you to expose your own tools — database queries, internal APIs, proprietary data sources — to agents without building custom integrations.

MCP connections are configured at the agent level and scoped per-tenant for security isolation.

Tool security principles

Least privilege — only enable tools the agent actually uses
No credentials in SOUL files — use environment variables for API keys
Audit tool calls — the activity log records every tool invocation with timestamps
Review regularly — as agents evolve, remove tools that are no longer needed

See Access Control for per-agent permission management.