AI Extensions, October 30: package facts and cloud permissions need different checks
Anaconda’s hosted MCP supplies package intelligence without changing environments. Azure MCP’s current beta belongs behind a separate cloud-permission review.

Anaconda’s hosted MCP supplies package intelligence without changing environments. Azure MCP’s current beta belongs behind a separate cloud-permission review.
Anaconda: a hosted lookup, with a dated launch
The primary announcement dates general availability to October 2, not today. I revisited it because an October 30 secondary headline made the launch sound new. The current service is remote and read-only, unlike the older local beta. Its four tools cover package metadata, vulnerability information, organization configuration and forum search. An account is required; the current table puts package_security and org_config on Business or Enterprise tiers. I have not connected an account or checked a specific subscription.

The installation decision stays with the agent
The official CLI page says ana mcp setup configures selected assistants and installs the Anaconda Intelligence skill; restart the assistants afterward. ana mcp clients checks default connection entries, not actual connectivity or custom names. The remote service cannot install packages, and it does not enforce policy on other tools. A package being present on a channel is not approval to install it. Review the package, version, platform and applicable policy before any environment change.
A setup trap in the older documentation
An older getting-started page still describes the local anaconda-mcp beta. The GA announcement recommends removing that beta package. Follow the current hosted-service documentation rather than mixing both recipes. This is a hosted commercial service review, not a claim that server code has an open-source license. No package installation, credential exchange or environment mutation was performed for this digest.
Azure MCP: a beta version is a deployment choice
The release list currently includes Azure.Mcp.Server 4.0.0-beta.4 dated October 29. Its newest release description still contains template placeholders, so I will not invent a feature summary from the version number. The README also discusses 2.0 GA; that statement does not make the 4.0 beta stable. The repository LICENSE is MIT. The README offers IDE installation and manual client setup, including npx @azure/mcp@latest server start. That moving tag is a documented example, not my recommendation for a reproducible rollout.
Review the tool surface before the cloud account
My proposed trial pins a reviewed version, uses an isolated subscription and the minimum permitted role, and checks which tools are exposed before asking for any operation. Authentication and cloud authorization must be reviewed separately from the license. Do not assume an Azure server is globally read-only because Anaconda’s service is. This digest is source review, not an installation or security audit.
A useful acceptance record
For each extension, record its source URL, checked date, version or hosted endpoint, permitted reads, possible writes, account requirements and disconnection route. A successful settings entry is not a successful connection test. Keep an authorized read-only query and an explicit refusal test for a forbidden change as separate checks. Popularity counts are omitted because they cannot establish permission or fit.
Sources
- https://www.anaconda.com/blog/anaconda-mcp-general-availability
- https://www.anaconda.com/docs/cli-reference/anaconda-mcp/main
- https://www.anaconda.com/docs/cli-reference/anaconda-cli/mcp
- https://github.com/microsoft/mcp/releases
- https://raw.githubusercontent.com/microsoft/mcp/main/servers/Azure.Mcp.Server/README.md
- https://raw.githubusercontent.com/microsoft/mcp/main/LICENSE