Cursor Cloud supplement

The rules below apply only when running inside a Cursor Cloud VM. The cloud VM update script appends this file to AGENTS.md at provisioning time, so the rest of the agent context is the same as a local checkout. Local developers and other agents should ignore this section — it is not relevant outside the cloud VM. Anything that already lives in AGENTS.md must not be repeated here; only cloud-VM-specific bootstrap, credentials, and artifact-handling rules belong below.

Cloud VM tooling

Repeatable installs (Docker, Node, npm dependencies, agent-browser, AWS CLI, Playwright Chromium) belong in the Cursor Cloud VM update script, not in this file. The cloud image additionally provides:

• agent-browser — browser automation for agents without a desktop.

• AWS CLI — uploads non-sensitive reproduction artifacts to AWS_S3_BUCKET_NAME.

Running Mattermost + plugin locally in the cloud VM

A local Mattermost Enterprise instance is needed for plugin development. Use the development image and a companion PostgreSQL container:

docker network create mm-network 2>/dev/null || true

docker run -d --name mm-postgres --network mm-network \
  -e POSTGRES_USER=mmuser -e POSTGRES_PASSWORD=mostest -e POSTGRES_DB=mattermost \
  -p 5432:5432 postgres:15-alpine

docker run -d --name mm-server --network mm-network -p 8065:8065 \
  -e MM_SQLSETTINGS_DRIVERNAME=postgres \
  -e MM_SQLSETTINGS_DATASOURCE="postgres://mmuser:mostest@mm-postgres:5432/mattermost?sslmode=disable&connect_timeout=10" \
  -e MM_SERVICESETTINGS_SITEURL=http://localhost:8065 \
  -e MM_PLUGINSETTINGS_ENABLEUPLOADS=true \
  -e MM_PLUGINSETTINGS_ENABLEUPLOAD=true \
  -e MM_PLUGINSETTINGS_AUTOMATICPREPACKAGEDPLUGINS=false \
  -e MM_PLUGINSETTINGS_ENABLEMARKETPLACE=false \
  -e MM_SERVICESETTINGS_ENABLELOCALMODE=true \
  mattermostdevelopment/mattermost-enterprise-edition:master

docker exec mm-server mmctl user create --email admin@example.com --username admin --password 'Admin1234!' --system-admin --local
docker exec mm-server mmctl team create --name dev --display-name "Dev Team" --local
docker exec mm-server mmctl team users add dev admin --local

The development image provides an Entry license for local testing; MM_LICENSE is not required.

Deploying the plugin into the cloud VM container

rm -rf server/dist dist
make dist-ci
MM_SERVICESETTINGS_SITEURL=http://localhost:8065 \
  MM_ADMIN_USERNAME=admin MM_ADMIN_PASSWORD='Admin1234!' \
  ./build/bin/pluginctl deploy mattermost-ai dist/*.tar.gz

Always clean server/dist first — a previous full make dist may have left other-platform binaries that bloat the upload.

Configuring an Anthropic agent via API

After deploying the plugin and starting Mattermost, configure an Anthropic service and create a user agent using the ANTHROPIC_API_KEY environment variable. Do not use /api/v4/config/patch for plugin config — write to PUT /plugins/mattermost-ai/admin/config.

curl -s -X POST http://localhost:8065/api/v4/users/login \
  -H 'Content-Type: application/json' \
  -d '{"login_id":"admin","password":"Admin1234!"}' \
  -D /tmp/mm-login-headers -o /tmp/mm-login-body >/dev/null
TOKEN=$(awk 'tolower($1)=="token:" {gsub("\r", "", $2); print $2}' /tmp/mm-login-headers)

curl -s -H "Authorization: Bearer $TOKEN" http://localhost:8065/plugins/mattermost-ai/admin/config |
  jq --arg key "$ANTHROPIC_API_KEY" '
    .services=[{
      id:"anthropic-service",
      name:"Anthropic",
      type:"anthropic",
      apiKey:$key,
      defaultModel:"claude-sonnet-4-5-20250929",
      tokenLimit:200000,
      outputTokenLimit:4096,
      streamingTimeoutSeconds:120
    }] |
    .bots=[] |
    .defaultBotName="" |
    .enableChannelMentionToolCalling=true |
    .mcp.enabled=true |
    .mcp.embeddedServer.enabled=true |
    .mcp.enablePluginServer=true |
    .mcp.servers=[] |
    .mcp.idleTimeoutMinutes=30' >/tmp/agents-config.json

curl -s -X PUT http://localhost:8065/plugins/mattermost-ai/admin/config \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  --data-binary @/tmp/agents-config.json

curl -s -X POST http://localhost:8065/plugins/mattermost-ai/agents \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"displayName":"Claude Agent","username":"claudeagent","serviceID":"anthropic-service","customInstructions":"You are a concise helpful assistant.","channelAccessLevel":0,"userAccessLevel":0,"model":"claude-sonnet-4-5-20250929","enableVision":true,"disableTools":true}'

After this, @claudeagent is available for DMs and mentions. Config schema lives in config/config.go and llm/configuration.go; the agent API request shape is CreateAgentRequest in api/api_agents.go.

Verify embedded MCP tools are discoverable: GET /plugins/mattermost-ai/admin/mcp/tools (admin auth required).

Uploading PR artifacts to S3

All screenshots and walkthrough videos captured during development must be uploaded to S3 and linked in the PR description so reviewers see visual evidence of changes. Never include secrets, credentials, or API keys in screenshots or videos — retake or redact instead.

Required env vars (provided as cloud secrets): AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_S3_BUCKET_NAME.

BRANCH=$(git rev-parse --abbrev-ref HEAD)
PREFIX="cursor/$BRANCH"
aws s3 cp /opt/cursor/artifacts/ "s3://$AWS_S3_BUCKET_NAME/$PREFIX/" --recursive
REGION=$(aws s3api get-bucket-location --bucket "$AWS_S3_BUCKET_NAME" --output text)
[ "$REGION" = "None" ] || [ "$REGION" = "null" ] && REGION=us-east-1
for f in /opt/cursor/artifacts/*; do
  FILENAME=$(basename "$f")
  if [ "$REGION" = "us-east-1" ]; then
    URL="https://$AWS_S3_BUCKET_NAME.s3.amazonaws.com/$PREFIX/$FILENAME"
  else
    URL="https://$AWS_S3_BUCKET_NAME.s3.$REGION.amazonaws.com/$PREFIX/$FILENAME"
  fi
  case "$FILENAME" in
    *.gif|*.jpg|*.jpeg|*.png|*.webp) echo "![${FILENAME}](${URL})" ;;
    *) echo "[${FILENAME}](${URL})" ;;
  esac
done

Add image artifacts to the PR description as Markdown images (![alt](url)) so they render inline. Use plain links for non-image artifacts. Upload before creating or updating the PR description.

Cloud-VM-specific gotchas

• The Mattermost server port comes from the Docker port mapping above (8065); if you re-run with a different -p flag, look it up with docker port mm-server.

• Outbound LLM traffic from the container respects HTTP_PROXY / HTTPS_PROXY on the Mattermost process environment if the cloud VM routes egress through a proxy.