tiger-salesforce

Tiger Salesforce MCP Server wraps a TimescaleDB-backed Salesforce data store and exposes focused tools for LLMs via the Model Context Protocol. It provides a semantic search tool over Salesforce case summaries, enabling retrieval of relevant case narratives with embedding-based similarity. The server runs as a Node.js process and serves the MCP endpoints alongside REST APIs; it is designed to be consumed by MCP-compatible clients or by tools like Claude or Claude Desktop where you configure a local mcpServers entry to bridge between your client and the server. You can invoke semanticSearchSalesforceCaseSummaries to fetch top matching case summaries by a textual prompt, with an optional limit parameter to cap results.

To use it from an MCP client, point your MCP configuration to the node process (stdio transport) and provide the target PostgreSQL database credentials and an API key for the OpenAI service. The server reads your database connection details to fetch precomputed summaries and embeddings, then returns results containing case IDs, summaries, and a distance score indicating semantic closeness.

Prerequisites:

• Node.js (recommended LTS) installed on your machine or deployment environment
• Access to the TimescaleDB instance hosting the Salesforce-related data
• An OpenAI API key (or compatible LLM provider) for embedding and summarization capabilities

1. Clone the repository

git clone git@github.com:timescale/tiger-salesforce-mcp-server.git cd tiger-salesforce-mcp-server

2. Install dependencies

npm i

3. Configure environment

• Create a local .env file or set environment variables as shown in the example. At minimum provide database connection details and an API key.

cp .env.sample .env

4. Build (if applicable) and run

npm run build node dist/index.js

5. Optional: configure MCP client

• Use the provided example in Claude Desktop or your MCP client to connect to the server via stdio transport, providing the path to dist/index.js and ensuring the same environment variables are available in the client configuration.

Tips and common considerations:

• Ensure the PostgreSQL host, port, database, and credentials match what is configured for your TimescaleDB instance. The server uses these to read pre-generated Salesforce summaries and embeddings.
• Protect sensitive credentials (PGUSER, PGPASSWORD, OPENAI_API_KEY). Use secret management or Kubernetes Secrets in production.
• If you see connection or timeout errors, verify network access to the TimescaleDB instance and confirm that the MCP client transport (stdio) is correctly configured.
• When deploying to Kubernetes, consider mounting environment variables via a secret store and updating the mcpServers configuration accordingly.
• The semanticSearchSalesforceCaseSummaries tool supports a limit parameter; if omitted, a default of 10 results will be returned.