Integrate Long-Term Memory with Claude Code via MCP

Prerequisites

• Claude Code CLI is installed and functional.

• Python >= 3.10.

• uv package manager (for on-demand startup via uvx, recommended).

• Obtained the AnalyticDB for PostgreSQLlong-term memory service API URL and token. For details, seeContext service.

• Obtained the adbpgmem_mcp_selfhosted project source code adbpgmem-mcp-server-0.2.0.tar.gz.

  Extract the archive to a directory named adbpgmem_mcp_selfhosted. All paths in the following steps refer to this extracted directory.

Installation

No pre-installation is required for MCP Server. uvx creates a temporary virtual environment from the local project path on demand and starts the service. You only need to ensure that the source path is accessible.

# Verify the source path is accessible
ls /path/to/adbpgmem_mcp_selfhosted/pyproject.toml
# The file info should be displayed, confirming the path is correct

Parameters

Method 1: Inject via --env During Registration (Recommended)

You can inject environment variables directly through the --env parameter when registering the MCP Server, without creating additional configuration files. For more information, see the Register MCP Server section below.

Method 2: Manually Create a Configuration File

Global configuration (applies to all projects):

mkdir -p ~/.claude
cat > ~/.claude/adbpgmem.conf << 'EOF'
ADBPGMEM_API_URL="https://api-longmemory-cn-chengdu.opentrust.net"
ADBPGMEM_API_TOKEN="sk-your-token-here"
ADBPGMEM_USER_ID="your.username"
EOF

Project-level configuration (applies to the current project only):

mkdir -p .claude
cat > .claude/adbpgmem.conf << 'EOF'
ADBPGMEM_API_URL="https://api-longmemory-cn-chengdu.opentrust.net"
ADBPGMEM_API_TOKEN="sk-your-token-here"
ADBPGMEM_USER_ID="your.username"
EOF

Configuration File Description

The configuration file uses the KEY="VALUE" format, with one entry per line. If the configuration is missing, MCP Server logs a warning and continues to start, but API calls will fail.

Global Registration (Recommended)

Writes to ~/.claude.json and takes effect for all projects:

claude mcp add --scope user --transport stdio adbpgmem \
  --env ADBPGMEM_API_URL=https://api-longmemory-cn-chengdu.opentrust.net \
  --env ADBPGMEM_API_TOKEN=sk-your-token-here \
  --env ADBPGMEM_USER_ID=your.username \
  -- uvx --from /path/to/adbpgmem_mcp_selfhosted adbpgmem-mcp-server

The following table describes each parameter in the command:

Project-Level Registration

You can also register the MCP Server through the .mcp.json file in the project root directory. This method takes effect only for the current project and can be committed to version control for team sharing:

{
  "mcpServers": {
    "adbpgmem": {
      "command": "uvx",
      "args": ["--from", "/path/to/adbpgmem_mcp_selfhosted", "adbpgmem-mcp-server"],
      "env": {
        "ADBPGMEM_API_URL": "https://api-longmemory-cn-chengdu.opentrust.net",
        "ADBPGMEM_API_TOKEN": "sk-your-token-here",
        "ADBPGMEM_USER_ID": "your.username"
      }
    }
  }
}

Verify Registration

Check if the MCP Server is registered successfully:

# Check if the MCP Server is registered successfully
claude mcp list
# The output should include adbpgmem

# After starting Claude Code, type /mcp to check if the status shows ✓ Connected

Invocation

Claude autonomously calls MCP tools based on the conversation context. You trigger them through natural language:

Detailed Tool Usage

You: Remember: this project uses Go 1.22 + Gin framework, ORM uses GORM v2, database is PostgreSQL 15

Claude calls add_memory → Server stores the memory

You: Search for previous memories about database selection

Claude calls search_memories → Returns semantically matched historical memories

Scenario 1: Search Historical Debugging Experience

You encounter a Docker container out-of-memory issue and want to check whether there is any previous experience.

Operation:

You: Search if we've encountered Docker container OOM issues before

Claude behavior:

1. Recognizes the search intent.

2. Calls search_memories(query="Docker container OOM out of memory").

3. Returns matching historical memories.

Example output:

Found 1 relevant memory:

1. [score: 0.91] Java applications in Docker containers require -XX:MaxRAMPercentage=75.0
   and -XX:+UseContainerSupport, otherwise JVM cannot correctly recognize container
   memory limits, causing it to exceed the container limit and get killed.

Scenario 2: Cross-Session Knowledge Reuse

In a previous session, you established API design conventions. This session requires developing new API endpoints.

Session 1:

You: Remember: our APIs uniformly use RESTful style, pagination uses page/page_size,
    sorting uses sort=field:asc|desc, error responses include type/title/status/detail fields

Session 2 (days later, a new session):

You: Help me design a new user management API endpoint

Claude: Based on the API design conventions saved earlier, I'll design the user management endpoint:
       - List API supports pagination and sorting
       - Error responses use the unified four-field format
       - Follows RESTful naming conventions

       [Automatically outputs design following team conventions without needing repeated instructions]

Claude automatically retrieves the conventions established in the previous session from the memory store. New sessions can directly reuse team consensus without repeated communication.