AgentCrew

AgentCrew is a Python multi-agent AI framework and desktop/CLI application for building teams of specialized agents that collaborate via transfer or parallel delegation. It supports multiple model providers and tool integrations, including MCP-connected external tools, and can expose agents as HTTP services via an A2A (JSON-RPC) server.

Evaluated Mar 30, 2026 (21d ago)
Homepage ↗ Repo ↗ Ai Ml ai-ml ai-agents multi-agent-systems mcp tool-use python cli desktop-gui json-rpc a2a
⚙ Agent Friendliness
50
/ 100
Can an agent use this?
🔒 Security
41
/ 100
Is it safe for agents?
⚡ Reliability
38
/ 100
Does it work consistently?

Score Breakdown

⚙ Agent Friendliness

MCP Quality
45
Documentation
70
Error Messages
0
Auth Simplicity
55
Rate Limits
25

🔒 Security

TLS Enforcement
40
Auth Strength
40
Scope Granularity
20
Dep. Hygiene
55
Secret Handling
55

Strengths indicated: README mentions approval/denial before tool execution and configurable rate limits/access controls; credentials are stored in local files (e.g., ~/.codex/auth.json for Codex) implying local secret storage rather than a hosted service. Gaps/risks from provided material: no documented transport/security requirements for the A2A server (TLS/authentication details are unclear), and no evidence of fine-grained authorization scopes per tool/action. Dependency list is large and multi-provider; without vulnerability reports or lockfile/CVE data, hygiene is estimated.

⚡ Reliability

Uptime/SLA
0
Version Stability
50
Breaking Changes
40
Error Recovery
60
AF Security Reliability

Best When

You want a locally run multi-agent “agent team” that can use tools (including MCP) and optionally expose agent capabilities to other clients for orchestration.

Avoid When

You need a minimal surface area package with strong, documented security primitives (mTLS, JWT/OAuth with scopes on every endpoint) and formally specified HTTP API contracts.

Use Cases

  • Building specialized multi-agent assistants (research, coding, writing, architecture) that collaborate
  • Automating single-turn tasks with validated structured (JSON Schema) output in job mode
  • Connecting agents to external capabilities via Model Context Protocol (MCP) and other tool integrations (web search, file editing, command execution, etc.)
  • Running interactive chat sessions with multi-agent orchestration and approval gating for tool usage
  • Providing agent endpoints to other systems/instances via an A2A server

Not For

  • Production environments requiring strict, audited enterprise security controls without additional hardening
  • Use cases needing fine-grained authorization model for every tool/action with documented policy enforcement
  • Environments that cannot tolerate local execution of tools like shell commands or browser automation (even if approval is enabled)
  • Teams that require a stable, formally versioned public API contract with OpenAPI/grpc schemas

Interface

REST API
No
GraphQL
No
gRPC
No
MCP Server
No
SDK
No
Webhooks
No

Authentication

Methods: API keys for AI providers (e.g., Anthropic, OpenAI, etc.) via configuration files OAuth flows for specific integrations (GitHub Copilot OAuth, ChatGPT/Codex OAuth flow) MCP OAuth support (described as supported for MCP integration)
OAuth: Yes Scopes: No

Auth for agent execution is primarily via local configuration of provider credentials. The A2A server mode is described as exposing JSON-RPC endpoints and an /.well-known/agent.json discovery document, but explicit server-side auth requirements (e.g., API keys/JWT) and scope granularity are not documented in the provided README excerpt.

Pricing

Free tier: No
Requires CC: No

The framework is open-source (Apache-2.0). Ongoing spend is driven by underlying LLM/provider APIs and any hosted dependencies.

Agent Metadata

Pagination
none
Idempotent
False
Retry Guidance
Documented

Known Gotchas

  • Job mode retries up to 4 times for JSON Schema validation failures, which may increase cost/latency if schemas are strict or frequently mismatched.
  • Delegate mode runs multiple tool calls concurrently via asyncio.gather; tool failures are isolated per-tool but concurrency can make non-deterministic ordering/logging effects possible.
  • Tool integrations include potentially destructive actions (e.g., command execution, file editing); correctness and safety depend on configured permissions/approval gating.

Alternatives

Full Evaluation Report

Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for AgentCrew.

AI-powered analysis · PDF + markdown · Delivered within 30 minutes

$99

Package Brief

Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.

Delivered within 10 minutes

$3

Score Monitoring

Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.

Continuous monitoring

$3/mo

Scores are editorial opinions as of 2026-03-30.

8642
Packages Evaluated
17761
Need Evaluation
586
Need Re-evaluation
Community Powered