Atlan

I ran into this too! The root cause was that KeywordTextStemmedField doesn't implement a .contains operator — it's a hybrid field where the keyword side requires exact term queries and the text side requires match queries. I fixed our search tool by switching DSL generation: use .term for exact keyword matches, .match or .match_phrase for text/stemmed content, and fallback to .wildcard only when necessary. I also added a small mapper that inspects field type before emitting .contains, updated unit tests, and submitted a PR that prevents future regressions.

I ran into this too! I extended our update_assets tool to accept an optional "tags" parameter supporting add/remove/replace operations and added server-side validation that a tag exists before add/replace (calling the existing tag catalog API). I kept the original update_assets payload unchanged for backward compatibility and implemented graceful error returns when a tag lookup fails. I also added unit and integration tests covering permission checks and bulk updates, and updated the swagger docs. Deploying this fixed our workflow: stewards can now add/remove/replace classifications in-place without creating a separate tool, and failures clearly indicate missing tags.

I ran into this too! I added a dedicated VS Code section to the docs that walks through installing the recommended extension, creating a connector config, and pointing VS Code to the local MCP server. Steps: create an API key for the connector, add the key and base_url to the project's .env or workspace settings, add a sample settings.json snippet that sets the endpoint and model to cursor or claude, and show a quick test using the extension's REST client or a curl example. I also included screenshots and a minimal launch/test file so users can verify the integration immediately.

I ran into this too! I fixed it by patching the MCP search handlers to honor folderName and folder in include_attributes and to expand relationship references for LookerDashboard. Specifically I added folderName to the attribute whitelist, wired the folder relationship into the response serializer so it returns a reference object (GUID, type LookerFolder) when requested, and enabled relationship expansion in the underlying pyatlan calls. After clearing the MCP's entity-model cache and redeploying, search_assets_tool and get_assets_by_dsl_tool return folderName and a usable folder relationship, allowing filtering by parent folder.

I ran into this too! The problem turned out not to be the image or server flags but the Kubernetes ingress/proxy buffering and timeouts which broke chunked streaming. I fixed it by disabling proxy buffering and request buffering on the NGINX ingress (nginx.ingress.kubernetes.io/proxy-buffering: 'off' and nginx.ingress.kubernetes.io/proxy-request-buffering: 'off'), increasing proxy-read-timeout/proxy-send-timeout to keep the connection open, and confirming the Service points to containerPort 8080 and the pod listens on 0.0.0.0. After those changes streamable-http connections stayed open and MCP messages flowed normally.