Agent Harness Integrations

Remember connects seamlessly to modern AI coding agents via the Model Context Protocol (MCP).

With Remember, your coding agent gains:

Persistent memory across sessions: Decisions, architectural invariants, and dependency rules persist even when context windows reset.
Auditable provenance: Every recalled fact links directly to the source document, PR, or discussion where it was asserted.
Contradiction protection: When requirements change, Remember invalidates old facts and flags active contradictions instead of hallucinating a blended answer.

1. Automatic Setup with `remember setup`

The easiest way to configure any harness is running remember setup in your repository root:

uvx remember setup

The bootstrapper detects installed agent tools in your environment and workspace, registers the MCP server with durable absolute binary paths, and generates tailored context rules.

2. Cursor Integration

Automatic Configuration

remember setup creates two files in your workspace:

.cursor/mcp.json: Registers the Remember MCP server.
.cursor/rules/remember.mdc: Guides Cursor on when and how to query memory during chat and composer runs.

Manual Configuration

Add this entry to your workspace or global .cursor/mcp.json:

{
  "mcpServers": {
    "remember": {
      "command": "uvx",
      "args": ["remember", "mcp"]
    }
  }
}

Zero Secret Leakage: The MCP server automatically loads credentials from ~/.config/remember/credentials.json (mode 0600). Bearer tokens are never written to repository files committed to git.

(If self-hosting locally, add "env": { "REMEMBER_DATA_PLANE_URL": "http://localhost:8000" })

Rule File (`.cursor/rules/remember.mdc`)

Place this in .cursor/rules/remember.mdc to guide Cursor:

---
description: Persistent architectural memory and decision log
globs: *
alwaysApply: true
---
 
# Remember Memory Instructions
 
You have access to Remember via MCP tools.
- Before refactoring or making architectural assumptions, call `facts_context` or `combined_context` to check for existing decisions and invariants.
- If a query returns contradictions (`contradictions` array in the response), disclose them explicitly.
- For deep structural exploration, use `query_sql` against `facts_current` or `graph_edges_current`.

3. Claude Code Integration

Quick Registration

Run the Claude Code CLI command:

claude mcp add remember -- uvx remember mcp

Self-Hosted Local Configuration

claude mcp add remember \
  -e REMEMBER_DATA_PLANE_URL="http://localhost:8000" \
  -- uvx remember mcp

Now whenever you launch claude, Claude Code has access to Remember's tools across all tasks in your repository.

4. Claude Desktop Integration

Configuration File

Open your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json

Add the remember server to mcpServers:

{
  "mcpServers": {
    "remember": {
      "command": "uvx",
      "args": ["remember", "mcp"]
    }
  }
}

Restart Claude Desktop. You will see a hammer icon indicating Remember's memory tools are active.

5. OpenAI Codex Integration

Add the MCP server configuration to your workspace .codex/config.toml or user configuration ~/.codex/config.toml:

[mcp_servers.remember]
command = "uvx"
args = ["remember", "mcp"]

6. Google Antigravity (AGY) Integration

remember setup creates .agents/mcp_config.json and a specialized skill at .agents/skills/remember/SKILL.md:

{
  "mcpServers": {
    "remember": {
      "command": "uvx",
      "args": ["remember", "mcp"]
    }
  }
}

The generated SKILL.md teaches Antigravity agents the Orient, Verify, Audit lifecycle:

Orient: Check high-level knowledge documents.
Verify: Call facts_context for load-bearing facts.
Audit: Trace claims and source citations before irreversible actions.

7. Exposed MCP Tools

When your agent connects to remember mcp, the following tools are advertised:

Tool Name	Intent	Description
`facts_context`	Fact Recall	Retrieves current or historical adjudicated facts with supporting evidence under a token budget.
`combined_context`	Comprehensive	Returns a dual-grain `ContextBundle` containing both raw testimony (claims) and adjudicated truth (facts).
`claims_and_sources_context`	Evidence Search	High-recall semantic retrieval over raw claims and chunk source text.
`resolve_entity`	Disambiguation	Resolves ambiguous entity names (e.g. "Postgres", "Alice") to canonical entity IDs.
`query_sql`	Open Query	Executes AST-sandboxed, read-only SQL queries against the `memory_v1` view schema.
`describe_query_space`	Schema Discovery	Returns the complete `memory_v1` relational views, columns, and graph helper signatures.
`search_query_space`	Discovery Search	Searches documentation and view descriptions across the query schema.

8. Verifying Your Setup

Whenever you change configuration or switch projects, run:

remember doctor

remember doctor verifies:

Token validity and permissions.
Data plane responsiveness (Cloud or local Docker Compose).
Configuration syntax across all detected agent harnesses.
Stdio MCP bootability without latency traps.