byted-volcengine-flink

Byted Volcengine Flink

Volcano Engine Flink aggregation skill entry, used to uniformly handle Flink-related requirements and route them to the corresponding sub-skill documents based on intent.

Prerequisites

Before installing and using this skill, you need to first complete the flink-mcp prerequisites.

1) Install and verify mcporter

flink-dev.md、flink-resource.md、flink-sre.md The O&M/diagnostic commands in depend on mcporter。

You can choose either installation method based on your Python environment:

npm install -g mcporter

Verify after installation:

mcporter --help

2) Configure the required environment variables

flink-mcp Startup depends on the following environment variables (all are required):

• VOLCENGINE_ACCESS_KEY
• VOLCENGINE_SECRET_KEY
• VOLCENGINE_REGION
• VOLCENGINE_PROJECT_NAME

Example:

export VOLCENGINE_ACCESS_KEY="your-access-key"
export VOLCENGINE_SECRET_KEY="your-secret-key"
export VOLCENGINE_REGION="cn-beijing"
export VOLCENGINE_PROJECT_NAME="your-flink-project"

3) Register flink-mcp in the local MCP configuration file (recommended)

Add flink-mcp to the configuration file used by the local MCP Client (usually mcp.json or an equivalent configuration file).

Note: The MCP Client will launch flink-mcpbased on this configuration, so this is the default primary path.

First ensure that uv / uvxis installed locally, then write the following configuration:

{
  "mcpServers": {
    "mcp-server-flink": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/volcengine/mcp-server#subdirectory=server/mcp_server_flink",
        "mcp-server-flink",
        "-t",
        "streamable-http"
      ],
      "env": {
        "VOLCENGINE_ACCESS_KEY": "${VOLCENGINE_ACCESS_KEY}",
        "VOLCENGINE_SECRET_KEY": "${VOLCENGINE_SECRET_KEY}",
        "VOLCENGINE_REGION": "${VOLCENGINE_REGION}",
        "VOLCENGINE_PROJECT_NAME": "${VOLCENGINE_PROJECT_NAME}",
        "UV_INDEX_URL": "https://mirrors.ivolces.com/pypi/simple/"
      }
    }
  }
}

4) Optional: Manually start flink-mcp (only for integration testing/troubleshooting)

When you need to independently verify whether the service can start, you can manually run the following in the terminal:

uvx --from git+https://github.com/volcengine/mcp-server#subdirectory=server/mcp_server_flink mcp-server-flink -t streamable-http

Sub-skill references and routing

Route tasks to the following sub-skills based on user requirements:

• Flink SQL development/deployment:flink-dev.md
• Read-only diagnostics (change operations prohibited):flink-diagnosis.md
• Resource and fault analysis:flink-resource.md
• SRE O&M changes (start/stop/restart/scaling/parameter modification):flink-sre.md

Routing rules

1. When the user says "create SQL / develop SQL / deploy SQL / debug SQL", prioritize using flink-dev.md。
2. When the user explicitly requests "read-only troubleshooting, no changes of any kind", you must use flink-diagnosis.md。
3. When the user asks about fault root causes, OOM, Checkpoint, performance issues, or connection issues, prioritize using flink-resource.md。
4. When the user requests O&M actions (start, stop, restart, scale up, scale down, modify configuration), you must use flink-sre.md, and perform a risk confirmation first.

Execution principles

• When information is insufficient, first fill in the key parameters: project name, task name, time range, target action.
• Any change-type operation (especially in flink-sre.md ) must first have the user confirm the risk.
• When the user only needs troubleshooting, adhere to a read-only toolchain and do not trigger actions such as start/stop/deploy.