3. Toolkits for third-party services and Cloud-Security composition¶

Part of the Exposure, Interoperability & Hardening set. Previous: A2A · Next: Interaction surface

This chapter is a curated catalogue of toolkits in packages/ai-parrot-tools/src/parrot_tools/ that are most useful as building blocks for vendor-specific MCP servers or for composing domain agents (e.g. cloud-security audit, ERP integration, finance research). Every toolkit inherits from AbstractToolkit (parrot_tools/toolkit.py); each public async method automatically becomes an agent tool through introspection — meaning a single MCPServer(toolkit=...) line yields a fully functional MCP service.

3.1 Composition map¶

graph LR
    subgraph Base["Base abstractions"]
        Abs["AbstractTool · ToolResult<br/>parrot_tools/abstract.py"]
        Tk["AbstractToolkit<br/>parrot_tools/toolkit.py"]
    end

    subgraph Enterprise["Enterprise SaaS"]
        Jira["JiraToolkit"]
        Odoo["OdooToolkit"]
        Workday["WorkdayToolkit"]
        Backstage["BackstageCatalogToolkit"]
        O365["SharePoint · OneDrive · Mail"]
        Teams["MSTeamsToolkit"]
        Google["Google Suite tools"]
    end

    subgraph Cloud["AWS"]
        S3["S3Toolkit"]
        IAM["IAMToolkit"]
        EC2["EC2Toolkit"]
        ECS["ECSToolkit"]
        EKS["EKSToolkit"]
        CW["CloudWatchToolkit"]
        Lambda["LambdaToolkit"]
        ECR["ECRToolkit"]
        Route53["Route53Toolkit"]
        RDS["RDSToolkit"]
        DDB["DocumentDBToolkit"]
        GD["GuardDutyToolkit"]
        SH["SecurityHubToolkit"]
    end

    subgraph Sec["Cloud-Security composition"]
        CloudSploit["CloudSploitToolkit"]
        Posture["CloudPostureToolkit<br/>(Prowler)"]
        Container["ContainerSecurityToolkit<br/>(Trivy)"]
        IaC["SecretsIaCToolkit<br/>(Checkov)"]
        Compliance["ComplianceReportToolkit<br/>multi-scanner aggregator"]
    end

    subgraph Data["Data & DBs"]
        DBQ["DatabaseQueryTool"]
        Multi["MultiTierSchemaCaching"]
        ES["ElasticsearchTool"]
        Arango["ArangoDBSearchTool"]
        DM["DatasetManager"]
    end

    subgraph Code["Code execution"]
        CI["CodeInterpreterTool"]
        Shell["ShellTool"]
        Sandbox["SandboxTool (gVisor)"]
        Docker["DockerToolkit"]
        Pulumi["PulumiToolkit"]
    end

    subgraph Web["Web & search"]
        Scrape["WebScrapingToolkit"]
        Site["SiteSearchToolkit"]
        Serp["SerpApiSearchTool"]
        Bing["BingSearchTool"]
        DDG["DuckDuckGoSearchTool"]
    end

    subgraph Finance["Finance / quant"]
        Quant["QuantToolkit"]
        Bloom["BloombergTool"]
        IBKR["IBKR toolkit"]
        YF["YFinanceTool"]
        FRED["FredApiTool"]
        Massive["MassiveToolkit"]
    end

    subgraph Docs["Documents"]
        PDF["PDFPrintTool"]
        Word["MSWordTool"]
        Excel["ExcelTool"]
        PPT["PowerPointTool"]
        FR["FileReaderTool"]
    end

    subgraph Output["Exposure"]
        MCP["MCP Server<br/>(any transport)"]
        A2A["A2A Server"]
        Agent["Agent / Crew"]
    end

    Abs --> Tk
    Tk --> Enterprise
    Tk --> Cloud
    Tk --> Sec
    Tk --> Data
    Tk --> Code
    Tk --> Web
    Tk --> Finance
    Tk --> Docs

    Cloud -. feeds .-> Sec
    Sec --> Compliance

    Enterprise --> MCP
    Enterprise --> A2A
    Enterprise --> Agent
    Cloud --> MCP
    Sec --> MCP
    Data --> Agent
    Code --> Agent
    Web --> Agent
    Finance --> Agent
    Docs --> Agent

    classDef base fill:#fff3e0,stroke:#ef6c00;
    classDef sec  fill:#fce4ec,stroke:#c2185b;
    classDef cld  fill:#e3f2fd,stroke:#1976d2;
    classDef out  fill:#e8f5e9,stroke:#2e7d32;
    class Abs,Tk base;
    class CloudSploit,Posture,Container,IaC,Compliance sec;
    class S3,IAM,EC2,ECS,EKS,CW,Lambda,ECR,Route53,RDS,DDB,GD,SH cld;
    class MCP,A2A,Agent out;

3.2 Atlassian / Jira¶

jiratoolkit.py → JiraToolkit — issues, projects, transitions over basic / token / OAuth1 auth.
parrot/auth/jira_oauth.py:86 → JiraOAuthManager — Atlassian 3LO with CSRF nonce, distributed token-refresh lock, cloud-id discovery, Redis-backed per-user token storage.

Recommended composition: MCPServer(transport="sse", auth="oauth2_external") + JiraToolkit → drop-in Atlassian MCP server.

3.3 Odoo ERP¶

odoo/toolkit.py → OdooToolkit — Odoo 14–19+, JSON-2 / XML-RPC auto-detection, bulk CRUD, external-id upsert via import_records, binary upload helper.

3.4 AWS — service-by-service¶

Toolkit	Class	Surface
`aws/s3.py`	`S3Toolkit`	Buckets, ACL/policy, encryption, public-access analysis.
`aws/iam.py`	`IAMToolkit`	Roles, users, policies, key audit, priv-esc detection.
`aws/ec2.py`	`EC2Toolkit`	Instance lifecycle, SGs, ENIs.
`aws/ecs.py`	`ECSToolkit`	Clusters, services, tasks.
`aws/eks.py`	`EKSToolkit`	EKS provisioning + management.
`aws/cloudwatch.py`	`CloudWatchToolkit`	Logs, metrics, alarms.
`aws/route53.py`	`Route53Toolkit`	DNS records, health checks.
`aws/rds.py`	`RDSToolkit`	Instances, snapshots, backups.
`aws/documentdb.py`	`DocumentDBToolkit`	DocumentDB management.
`aws/lambda_func.py`	`LambdaToolkit`	Functions, versions, aliases.
`aws/ecr.py`	`ECRToolkit`	Image registry + lifecycle.
`aws/guardduty.py`	`GuardDutyToolkit`	Threat findings + remediation.
`aws/securityhub.py`	`SecurityHubToolkit`	Aggregated findings + posture.

3.5 Cloud Security composition¶

The Cloud-Security pattern is the strongest illustration of toolkit composition. The five toolkits below cover scanning, IaC linting, container image analysis, and unified compliance reporting:

Toolkit	Class	Engines
`cloudsploit/toolkit.py`	`CloudSploitToolkit`	CloudSploit scanner orchestration.
`security/cloud_posture_toolkit.py`	`CloudPostureToolkit`	Prowler (AWS / Azure / GCP / K8s).
`security/container_security_toolkit.py`	`ContainerSecurityToolkit`	Trivy (images, FS, git, K8s, IaC).
`security/secrets_iac_toolkit.py`	`SecretsIaCToolkit`	Checkov (Terraform, CFN, K8s, Helm…).
`security/compliance_report_toolkit.py`	`ComplianceReportToolkit`	Multi-scanner aggregation + framework mapping.

A cloud-security agent is built by registering all five plus the relevant aws/* toolkits, exposing the composite as an MCP server, and gating the destructive actions through PBAC (chapter 5).

3.6 Microsoft 365 and Google¶

MS Teams: msteams.py → MSTeamsToolkit (messages, adaptive cards, meetings via Graph API).
O365 / SharePoint / OneDrive: o365/bundle.py → SharePointToolkit, OneDriveToolkit, o365/mail.py.
Google: googlesearch.py, googlelocation.py, googleroutes.py, googlesitesearch.py, googlevoice.py, all sharing google/base.py:GoogleToolArgsSchema + GoogleAuthMode (service account / user / cached).

3.7 Database & data¶

database/ + databasequery.py → DatabaseQueryTool (multi-DB SQL).
multidb.py → MultiTierSchemaCaching (in-memory → vector → live).
elasticsearch.py, arangodbsearch.py, querytoolkit.py.
dataset_manager/ → multi-source dataset loading.

3.8 Code execution and sandboxing¶

codeinterpreter/tool.py → CodeInterpreterTool (analysis, doc-gen, test generation).
shell_tool/tool.py → ShellTool — interactive PTY shell with plan-mode DAG + parallel command orchestration.
sandboxtool.py → SandboxTool — gVisor (runsc) kernel-level isolation for LLM-generated code.
docker/toolkit.py → DockerToolkit (container lifecycle + Compose).
pulumi/toolkit.py → PulumiToolkit (IaC plan / apply / destroy).

3.9 Web & scraping¶

scraping/toolkit.py → WebScrapingToolkit (Playwright/Selenium with AI-driven extraction).
sitesearch/toolkit.py, webapp_tool.py, serpapi.py, bingsearch.py, googlesearch.py, ddgsearch.py.

3.10 Messaging, finance, HR and documents¶

Domain	Toolkits
Messaging	`messaging/whatsapp.py`, `notification.py`, `zoomtoolkit.py`.
HR / ERP	`workday/tool.py` (`WorkdayToolkit`, OAuth2 + Redis cache, multi-WSDL).
Finance	`quant/toolkit.py` (`QuantToolkit`), `bloomberg.py`, `ibkr/`, `yfinance.py`, `fred_api.py`, `massive/`.
Documents	`pdfprint.py`, `msword.py`, `excel.py`, `powerpoint.py`, `doc_converter.py`, `file/`, `file_reader.py`.
DevTools	`gittoolkit.py` (`GitToolkit`), `backstage/toolkit.py` (`BackstageCatalogToolkit`), `flowtask/tool.py`.
Navigator	`navigator/toolkit.py` → `NavigatorToolkit` (Programs, Modules, Dashboards, Widgets — extends Postgres).

3.11 Recipe — composing a vendor MCP server¶

from parrot.mcp.server import MCPServer
from parrot_tools.jiratoolkit import JiraToolkit

toolkit = JiraToolkit(host=..., username=..., api_token=...)

server = MCPServer(
    name="atlassian-mcp",
    transport="sse",                     # ChatGPT-compatible
    auth="oauth2_external",              # delegate to corporate IdP
    allowed_tools=["jira_search_issues", "jira_get_issue",
                   "jira_create_issue", "jira_transition"],
)
server.register_tools(toolkit.get_tools())
await server.serve(host="0.0.0.0", port=8765)

The exact same shape works for Odoo, AWS, CloudSploit, Workday or any combination thereof.