Read-Only Mode

Not every tool or team member needs to write to the knowledge base. Some should just benefit from what's already there.

When to use read-only

• New team members — their AI tools immediately have access to accumulated knowledge, without contributing until they're comfortable
• Evaluation — trying pgmemory before committing? Connect read-only and see what the retrieval quality looks like
• Security-sensitive contexts — some environments have policies about what AI tools can store
• Non-engineering tools — a PM's AI assistant can search the technical knowledge base without writing back
• Cross-team consumers — teams that want to consume another team's knowledge without mixing their own context in

Setup

{
  "mcpServers": {
    "pgmemory": {
      "command": "pgmemory",
      "args": ["mcp", "--read-only"]
    }
  }
}

The --read-only flag restricts the MCP server to search-only tools. Write operations (memory_store, source_ingest, etc.) are disabled.

What the tool sees

When the AI searches the knowledge base, it gets formatted results:

[1] (source: claude-code, relevance: 0.87)
The payment service validates webhook signatures using HMAC-SHA256...

[2] (source: source:internal-wiki, relevance: 0.82)
Deployments to production require approval from the #releases channel...

Knowledge from proxy sessions, MCP writes, and ingested sources — all surfaced through a single search.

Participation spectrum

pgmemory supports a range of participation levels:

Level	How	Best for
Full (proxy)	Automatic capture + retrieval	Engineers using Claude Code who want zero-effort contribution
Full (MCP)	Agent-controlled search, store, and maintenance	Engineers using Cursor, Windsurf, custom tools
Read-only	Search only, no writes	New hires, evaluators, PMs, cross-team consumers
Isolated	Separate database	Anyone who needs a private knowledge store

There's no forced contribution. The value of the shared store is strong enough that most people opt into full participation voluntarily.