这是我们Salesforce数据库的包装器，包含嵌入的案例摘要。这通过模型上下文协议为大型语言模型提供了一些重点工具。

原始数据通过Fivetran连接从Salesforce获取。这会在TimescaleDB数据库中填充一个模式。一个独立进程生成支持案例的LLM摘要，然后嵌入这些摘要。该服务会搜索这些摘要。

API

所有方法都以MCP工具和REST API端点的形式公开。

Salesforce 案件摘要语义搜索

根据搜索提示的语义嵌入，搜索Salesforce案例摘要中的相关条目。

工具名称 ： semanticSearchSalesforceCaseSummaries

API 端点 ： GET /api/semantic-search/salesforce-case-summaries

输入

（REST API 使用查询参数）

输出

发展

克隆并本地运行服务器。

建筑

运行npm i安装依赖并构建项目。使用npm run watch来重建变更。

基于.env.sample文件创建一个.env文件。

测试

MCP检查员非常实用。

Tiger Salesforce MCP Server

A wrapper around our Salesforce database, which contains embedded case summaries. This provides some focused tools to LLMs via the Model Context Protocol.

The raw data is sourced from Salesforce via a Fivetran connection. This populates a schema in a TimescaleDB database. A separate process generates LLM summaries of the support cases, and then embeddings of those summaries. This service searches those summaries.

API

All methods are exposed as MCP tools and REST API endpoints.

Salesforce Case Summary Semantic Search

Searches the Salesforce case summaries for relevant entries based on a semantic embedding of the search prompt.

Tool name : semanticSearchSalesforceCaseSummaries

API endpoint : GET /api/semantic-search/salesforce-case-summaries

Input

(use query parameters for REST API)

{
  "prompt": "Why can't I connect to my database?",
  "limit": 10, // optional, default is 10
}

Output

{
  "results": [
    {
      "case_id": "500Nv000005HMfaIAG",
      "summary": "# Some content ...",
      "distance": 0.40739564321624144,
    },
    // more results...
  ],
}

Development

Cloning and running the server locally.

git clone [email protected]:timescale/tiger-salesforce-mcp-server.git

Building

Run npm i to install dependencies and build the project. Use npm run watch to rebuild on changes.

Create a .env file based on the .env.sample file.

cp .env.sample .env

Testing

The MCP Inspector is very handy.

npm run inspector

Testing in Claude Desktop

Create/edit the file ~/Library/Application Support/Claude/claude_desktop_config.json to add an entry like the following, making sure to use the absolute path to your local tiger-salesforce-mcp-server project, and real database credentials.

{
  "mcpServers": {
    "tiger-salesforce": {
      "command": "node",
      "args": [
        "/absolute/path/to/tiger-salesforce-mcp-server/dist/index.js",
        "stdio"
      ],
      "env": {
        "PGHOST": "x.y.tsdb.cloud.timescale.com",
        "PGDATABASE": "tsdb",
        "PGPORT": "32467",
        "PGUSER": "readonly_mcp_user",
        "PGPASSWORD": "abc123",
        "OPENAI_API_KEY": "sk-svcacct"
      }
    }
  }
}

Deployment

We use a Helm chart to deploy to Kubernetes. See the chart/ directory for details.

The service is accessible to other services in the cluster via the DNS name tiger-salesforce-mcp-server.savannah-system.svc.cluster.local.

Secrets