agent-memory-mcp

A memory, docs, and repo context layer for engineering agents.

agent-memory-mcp helps agents work with live engineering context, not just isolated notes. It combines typed memory, document retrieval, and repository-aware tools so Claude, Cursor, Codex, and other MCP clients can recall decisions, search runbooks, inspect project docs, and reuse operational knowledge across sessions.

It is designed for engineering workflows such as:

• DevOps and platform operations

• infrastructure changes and rollback planning

• runbooks, changelogs, RFCs, and postmortems

• project-level memory that stays attached to the repo

Who This Is For

• teams using AI agents on real codebases, docs, and operational workflows

• DevOps, platform, and infra engineers who need more than chat history

• projects that want local-first memory today and a shared service path later

Related MCP server: Sovereign Universal Memory MCP

Why Not Just A Memory Tool

Most memory MCP servers focus on "store a note, recall a note."

agent-memory-mcp is aimed at a wider engineering context layer:

• typed memory for decisions, facts, patterns, and working context

• RAG indexing for project docs, changelogs, and knowledge files

• repo/file tools for reading and searching allowed project paths

• local SQLite storage with stdio today and HTTP/JSON-RPC when you need to share it

This makes it a better fit when the agent needs to answer questions like:

• "Why did we disable HPA on this service?"

• "What changed recently that could explain this regression?"

• "Which runbook or RFC matches this incident?"

Table of Contents

• Who This Is For

• Why Not Just A Memory Tool

• Features

• What Improved For Users

• Start Local In 3 Minutes

• Local-Only Mode

• Index Your Repo In 2 Commands

• Turn It Into A Team Service Later

• Installation Options — Homebrew, binary, source, Docker

• CLI Mode

• MCP client configuration — Claude Desktop, Cursor, Codex

• Recommended Workflow Snippets — what to paste into CLAUDE.md / .cursorrules

• CLI commands

• MCP tools reference — and JSON examples in docs/MCP_TOOLS.md

• Configuration — env vars, hot-reload, indexing safety

• Security And Operations

• Architecture

• macOS service installation

• Troubleshooting / FAQ

• Development

Reference docs: HOOKS · MCP_TOOLS · SHARED_SERVICE · STEWARDSHIP · SEDIMENTATION · BACKUP_RESTORE · SECURITY · THREAT_MODEL · CONTRIBUTING · CHANGELOG

Features

• Automatic session capture — Claude Code hooks auto-capture knowledge at session end, save checkpoints before context compression, and compile pending summaries at session start

• Typed persistent memory with 4 types: episodic, semantic, procedural, working

• Hybrid retrieval that combines embeddings with keyword/BM25-like ranking

• RAG indexing for project docs, changelogs, and knowledge archives (enabled by default in stdio/CLI mode; disabled by default in the Homebrew service preset — see Installation Options)

• Repo-aware file tools for listing, reading, and searching allowlisted paths

• Knowledge stewardship — automated maintenance: duplicate detection, conflict resolution, stale detection, drift scanning, and a review inbox

• Temporal knowledge model — track when knowledge was valid, build supersession chains, and query "what was true at time T"

• Dual transport: stdio for MCP clients, HTTP/JSON-RPC for APIs and shared setups

• SQLite storage for both memory and vector index -- no external databases needed

• Auto-indexing with file watcher for long-running local or service mode

What Improved For Users

• Lower memory usage: memory store now reads from SQLite directly instead of loading everything into RAM — large memory banks no longer risk OOM

• Opinionated solo-local setup: one recommended layout, one data directory, one quick smoke path

• Auto-loaded .env: run from your project root without manually sourcing environment variables

• Local-only embedding mode: keep hosted providers disabled and send text only to your local Ollama endpoint

• Safer semantic recall: memories from a different embedding model no longer produce misleading matches

• Explicit migration flow: use agent-memory-mcp reembed for memory migration and agent-memory-mcp index for RAG rebuilds after switching models

• Better visibility: stats and memory_stats now show how many memories belong to each embedding model

• Ready MCP client configs: generate copy-paste snippets for Claude Desktop, Cursor, and Codex

• Safer indexing defaults: built-in directory excludes, optional per-path exclude globs, and secret redaction before documents are indexed

• Source-aware retrieval: docs, ADRs, RFCs, changelogs, runbooks, postmortems, CI configs, Helm, Terraform, and K8s files are classified and surfaced with source metadata

• Hybrid ranking for search: semantic similarity is now combined with keyword matches, recency, and source-aware weighting instead of cosine similarity alone

• Trust-aware retrieval: memory and document results now expose source_type, confidence, freshness, owner, and last_verified_at, and ranking uses trust/freshness instead of similarity alone

• Explainable retrieval: opt-in debug output shows filters, score components, and applied boosts for every result

• DevOps-first tools: store decisions, incidents, runbooks, and postmortems with domain-specific MCP tools instead of generic memory calls

• Memory lifecycle: memories move through statuses — active, outdated, superseded, canonical — so stale knowledge gets downranked automatically instead of polluting recall

• Manual consolidation workflow: merge duplicates, mark outdated notes, promote canonical entries, and inspect conflict groups without deleting history

• Explicit canonical knowledge layer: list and recall confirmed knowledge separately from raw memory, and surface canonical context first in project summaries

• Project bank views: see maintained knowledge organized by category — decisions, runbooks, incidents, caveats, migrations, review queue — instead of a flat memory list

• Session close pipeline: when a session ends, memory is analyzed, classified, and consolidated with existing knowledge instead of blindly appended

• Explainable consolidation: session close reports show what will be added, merged, outdated, or promoted, with a decision trace and risk level for each action

• DevOps session modes: close-session adapts behavior based on session type — incident and migration sessions get stricter review-first policy, coding sessions auto-apply low-risk updates

• Shared service packaging: a working Docker Compose recipe, shared env template, nginx reverse proxy example, and a dedicated shared deployment guide

• Built-in retrieval console: inspect hybrid ranking, trust, and normal-vs-debug retrieval in a lightweight HTTP UI at /console

• Safer HTTP defaults: HTTP mode binds to 127.0.0.1 by default; non-loopback binds require auth unless you explicitly opt into unsafe unauthenticated access

• Consistent CLI and MCP behavior: memory type validation, tag normalization, query/content limits, and trust summaries now follow the same policy across both interfaces

• Knowledge stewardship: steward_run executes a full maintenance cycle — duplicate detection, conflict resolution, stale entry scanning, and canonical promotion candidates — with a single command

• Stewardship inbox: review-required actions from maintenance runs, drift scans, and session consolidation land in one actionable queue instead of being silently applied or lost

• Drift detection: drift_scan compares memory entries against live repo files and docs to find stale, missing, or changed references

• Verification model: verify_entry and verification_candidates let agents and users track when knowledge was last verified and what needs attention

• Canonical health diagnostics: steward runs now include a health summary for canonical entries — stale, unverified, conflicting, and low-support

• Policy-governed automation: stewardship thresholds, auto-apply rules, and scheduling are configurable via steward_policy and environment variables

• Temporal knowledge: memories can carry valid_from / valid_until timestamps, and recall_as_of retrieves knowledge that was valid at a specific point in time

• Supersession chains: mark_outdated with a superseding entry automatically builds bidirectional links (superseded_by / replaces) and sets temporal boundaries

• Knowledge timeline: knowledge_timeline shows the chronological evolution of knowledge on a topic

• Age-aware recall: recall scoring applies a configurable exponential age decay so stale, non-evergreen memories sink in results while canonical knowledge and character/identity stay put — tune the half-life with MCP_RECALL_HALFLIFE_DAYS (default 30 days; 0 disables)

• Self-healing duplicate cleanup: the steward can auto-merge high-confidence, near-identical duplicate groups instead of only queuing them for review — opt-in and guarded by a content-similarity threshold so nothing unique is archived (auto_merge_duplicate_min_confidence in steward_policy)

• No more double session-close records: closing a task folds the auto-captured session summary into the finalize record instead of writing a second near-identical memory per slug, cutting the duplicate pairs the steward used to flag as false contradictions

Start Local In 3 Minutes

The recommended path is: run locally first, prove value on one repo, then expand.

Run these commands from your project root.

Prerequisites

Install the binary with one of these options:

# Homebrew (macOS/Linux) — recommended, auto-configures Claude Code hooks
brew tap ipiton/tap
brew install agent-memory-mcp

# go install
go install github.com/ipiton/agent-memory-mcp/cmd/agent-memory-mcp@latest

Then configure one embedding provider:

• Jina AI API key for the quickest hosted setup

• OpenAI API key or another OpenAI-compatible endpoint

• Ollama with bge-m3 for a local setup

1. Configure local mode

cp .env.example .env
# Edit .env:
# - keep the solo-local defaults unless you need to change them
# - enable at least one embedding provider
#   JINA_API_KEY, OPENAI_API_KEY, or OLLAMA_BASE_URL

The binary auto-loads .env from the current directory, so you do not need source .env.

The recommended solo-local preset keeps all runtime state inside one directory:

.agent-memory/
  rag-index/
  memory-store/
  logs/

Local-Only Mode

Use local-only mode when you want embeddings without sending text to hosted APIs.

cp .env.example .env
# Then set:
# MCP_EMBEDDING_MODE=local-only
# JINA_API_KEY=
# OPENAI_API_KEY=

In local-only mode:

• agent-memory-mcp never calls Jina AI

• agent-memory-mcp never calls OpenAI-compatible embedding APIs

• embeddings are generated only through a local backend: Ollama or llama.cpp

What still uses the network:

• the local Ollama HTTP endpoint, typically http://localhost:11434

• or a local llama.cpp server, typically http://127.0.0.1:8080/v1

If no local backend is running or no supported local model is available, embedding requests fail with a local-only specific error telling you to start the backend or disable MCP_EMBEDDING_MODE=local-only.

Alternative local backend: llama.cpp

If you already run llama.cpp (Apple Silicon native, GGUF models), point the server at its OpenAI-compatible /v1/embeddings endpoint instead of installing Ollama. It is opt-in — set LLAMACPP_BASE_URL to enable it. Once set it joins the fallback chain before Ollama (Jina → OpenAI → llama.cpp → Ollama) and works in local-only mode.

# Start llama.cpp with an embedding model
llama-server -m bge-m3.gguf --embedding --pooling cls -c 8192 -ub 8192

# Then configure the MCP server
LLAMACPP_BASE_URL=http://127.0.0.1:8080/v1
LLAMACPP_EMBEDDING_MODEL=bge-m3
MCP_EMBEDDING_MODE=local-only

llama.cpp returns the model's native embedding dimension, so make sure MCP_EMBEDDING_DIMENSION matches it (1024 for bge-m3) — a mismatch is rejected at recall time.

On slow self-hosted hardware (Ollama with bge-m3 on a low-core or ARM VPS), a single chunk can take 4-7 seconds to embed and the default 5s timeout will fire repeatedly. Raise the limits:

MCP_EMBEDDING_TIMEOUT=30s      # default 5s
MCP_EMBEDDING_MAX_RETRIES=3    # default 1

Invalid values fall back to the defaults, so the service still starts.

Concurrency tuning when auto-index / file watcher is on

A single-slot llama-server processes requests strictly serially. With MCP_RAG_AUTO_INDEX / MCP_RAG_FILE_WATCHER enabled, background reindex batches (50 chunks each) hold the only slot for tens of seconds, so interactive recall / semantic_search / index_documents queue behind them and hit context deadline exceeded — the server looks "degraded" even though throughput is fine. Give the embedding server parallel slots so interactive calls slip in alongside the batch:

llama-server -m bge-m3.gguf --embedding --pooling cls \
  -c 32768 -b 8192 -ub 8192 \   # 8192 ctx PER SLOT (see note) — fits the largest chunk
  -np 4 -cb \                   # 4 slots + continuous batching: interactive calls don't wait for the batch
  --metrics                     # exposes Prometheus /metrics; /slots shows live slot occupancy

-np splits the context. Per-slot context is ctx_size / n_parallel. bge-m3 is an encoder — every chunk must fit in one slot whole, and -b/-ub must be ≥ the largest chunk in tokens, or it fails with "input too large to process". So with -np 4 you need -c 32768 to keep 8192 per slot; do not lower -c, -b, or -ub below the single-slot value when adding slots.

Measured effect (Apple Silicon, bge-m3 Q8_0): a 50-input batch drops from ~50s to ~5s, and an interactive probe under batch load drops from 8–20s to ~0.03s.

Also smooth the reindex avalanche for large, frequently-edited files (whole-file re-chunk on every edit can re-trigger mid-cycle):

MCP_RAG_DEBOUNCE=2m         # default 30s — collapses bursts of edits into one reindex
MCP_RAG_WATCH_INTERVAL=5m   # periodic full-scan cadence

2. Start the local server

For MCP clients such as Claude Desktop, Cursor, or Codex:

agent-memory-mcp

For direct CLI use, the same binary already works without an MCP client:

agent-memory-mcp store -content "Ingress rollback uses previous Helm revision" -type procedural -tags "helm,rollback"
agent-memory-mcp recall "helm rollback"
agent-memory-mcp stats

3. Run a smoke check

agent-memory-mcp store -content "Solo local smoke check" -type working -tags "smoke,local"
agent-memory-mcp recall "solo local smoke"
agent-memory-mcp index
agent-memory-mcp search "agent memory"

If you are working from the source checkout, you can run the same flow with:

make local-smoke

Index Your Repo In 2 Commands

Once local mode is running against a project, index docs and search them:

agent-memory-mcp index
agent-memory-mcp search "recent ingress change"

Typical high-value sources include:

• docs/

• README.md

• CHANGELOG.md

• RFC / ADR folders

• runbooks and incident notes

Turn It Into A Team Service Later

When local mode proves useful, move in three steps:

1. solo local

2. team laptop with auto-indexing and file watching

3. shared service with HTTP mode, auth token, and reverse proxy

Fastest shared-service path:

cd deploy/docker
cp .env.shared.example .env.shared
# edit MCP_HTTP_AUTH_TOKEN and MCP_PROJECT_ROOT
docker compose --env-file .env.shared up -d --build

This keeps the same retrieval stack, but packages it for team use.

Reference docs:

• Shared Service Guide

• Security Policy

• Backup And Restore

Installation Options

Homebrew (recommended for macOS)

brew tap ipiton/tap
brew install agent-memory-mcp
brew services start agent-memory-mcp

This installs the binary, creates a default config, and starts the service on 127.0.0.1:18080 with memory enabled. RAG document search is disabled by default — enable it by editing the config:

# Edit config
nano $(brew --prefix)/etc/agent-memory-mcp/config.env

Set MCP_RAG_ENABLED=true, MCP_ROOT=/path/to/your/project, and MCP_INDEX_DIRS=docs,README.md. Changes are picked up automatically within ~30 seconds, or force reload with kill -HUP $(pgrep agent-memory-mcp).

Manage the service:

brew services restart agent-memory-mcp
brew services stop agent-memory-mcp
brew services info agent-memory-mcp

If you previously installed via Cask and want brew services:

brew uninstall --cask agent-memory-mcp
brew install ipiton/tap/agent-memory-mcp

Download a binary

Download a prebuilt archive from the Releases page.

The release archives include the version in their filename, so resolve the latest tag first:

# Resolve latest version once
VERSION=$(curl -fsSL https://api.github.com/repos/ipiton/agent-memory-mcp/releases/latest \
  | grep '"tag_name"' | head -1 | cut -d'"' -f4 | sed 's/^v//')

# macOS (Apple Silicon)
curl -fsSL "https://github.com/ipiton/agent-memory-mcp/releases/download/v${VERSION}/agent-memory-mcp-${VERSION}-darwin-arm64.tar.gz" | tar xz
sudo mv agent-memory-mcp /usr/local/bin/

# macOS (Intel)
curl -fsSL "https://github.com/ipiton/agent-memory-mcp/releases/download/v${VERSION}/agent-memory-mcp-${VERSION}-darwin-amd64.tar.gz" | tar xz
sudo mv agent-memory-mcp /usr/local/bin/

# Linux (x86_64)
curl -fsSL "https://github.com/ipiton/agent-memory-mcp/releases/download/v${VERSION}/agent-memory-mcp-${VERSION}-linux-amd64.tar.gz" | tar xz
sudo mv agent-memory-mcp /usr/local/bin/

# Linux (arm64)
curl -fsSL "https://github.com/ipiton/agent-memory-mcp/releases/download/v${VERSION}/agent-memory-mcp-${VERSION}-linux-arm64.tar.gz" | tar xz
sudo mv agent-memory-mcp /usr/local/bin/

Build from source

git clone https://github.com/ipiton/agent-memory-mcp.git
cd agent-memory-mcp
go build -o bin/agent-memory-mcp ./cmd/agent-memory-mcp

Docker

docker build -f deploy/docker/Dockerfile -t agent-memory-mcp .
docker run -p 18080:18080 \
  -v memory-data:/data \
  -e MCP_HTTP_MODE=http \
  -e MCP_HTTP_HOST=0.0.0.0 \
  -e MCP_HTTP_AUTH_TOKEN=replace-with-long-random-token \
  agent-memory-mcp

Or with docker compose:

cd deploy/docker
cp .env.shared.example .env.shared
docker compose --env-file .env.shared up -d --build

The MCP HTTP endpoint will be available at http://localhost:18080/mcp.

By default, bare-metal HTTP mode now binds to 127.0.0.1. For shared/container deployments, set MCP_HTTP_HOST=0.0.0.0 and a bearer token.

CLI Mode

The binary also works as a standalone CLI:

# Memory operations
agent-memory-mcp store -content "Project uses chi router" -type procedural -tags "go,chi"
agent-memory-mcp recall "router middleware"
agent-memory-mcp list -type procedural
agent-memory-mcp delete <memory-id>

# RAG search
agent-memory-mcp search "authentication flow"
agent-memory-mcp search -source-type runbook "ingress rollback"
agent-memory-mcp search -source-type runbook -debug "ingress rollback"
agent-memory-mcp index

# Project bank and session close
agent-memory-mcp project-bank canonical_overview
agent-memory-mcp close-session -summary "Updated payments rollback runbook after fixing ingress timeout" -context payments-api -service payments-api
agent-memory-mcp review-session -mode incident -stdin < notes/session.txt
agent-memory-mcp accept-session -summary "Added migration caveat for billing schema rename" -mode migration -context billing -service billing-api
agent-memory-mcp accept-session -raw-only -summary "Exploratory notes that are too noisy for consolidation"

# Utilities
agent-memory-mcp stats
agent-memory-mcp config claude-desktop
agent-memory-mcp reembed
agent-memory-mcp export > backup.json
agent-memory-mcp import backup.json

# JSON output for scripting
agent-memory-mcp recall "test" -json
agent-memory-mcp stats -json

Run agent-memory-mcp <command> -help for details on any command.

CLI memory commands and MCP memory tools now share the same validation and normalization rules:

• invalid memory types are rejected consistently

• comma-separated tags are trimmed and deduplicated the same way

• zero verified timestamps are hidden from trust summaries in both CLI and MCP output

When no command is given (or flags start with -), the binary starts the MCP server as before -- full backward compatibility.

MCP client configuration

Use the built-in generator to produce a project-local config that starts the server from your repo root.

This is the recommended path because it:

• keeps .env loading working without duplicating settings into every MCP client

• keeps .agent-memory/ relative to the project root

• gives you one copy-paste snippet per client

You can override the detected project root or binary path with -root and -command.

Claude Desktop

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json:

agent-memory-mcp config claude-desktop

Example generated output:

{
  "mcpServers": {
    "memory": {
      "command": "/bin/sh",
      "args": [
        "-lc",
        "cd '/path/to/your/project' && exec '/absolute/path/to/agent-memory-mcp'"
      ]
    }
  }
}

Cursor

Paste into ~/.cursor/mcp.json:

agent-memory-mcp config cursor

Example generated output:

{
  "mcpServers": {
    "memory": {
      "command": "/bin/sh",
      "args": [
        "-lc",
        "cd '/path/to/your/project' && exec '/absolute/path/to/agent-memory-mcp'"
      ]
    }
  }
}

Codex

Paste into ~/.codex/config.toml:

agent-memory-mcp config codex

Example generated output:

[mcp_servers.memory]
command = "/bin/sh"
args = ["-lc", "cd '/path/to/your/project' && exec '/absolute/path/to/agent-memory-mcp'"]

Rename the server or override paths

agent-memory-mcp config claude-desktop \
  -name engineering-memory \
  -root /path/to/your/project \
  -command /absolute/path/to/agent-memory-mcp

Recommended Workflow Snippets

Without these snippets, the agent will only use basic store_memory and recall_memory. To unlock session close, engineering memory types, project bank, and consolidation, add relevant snippets to your agent's instructions.

Where to put them:

• Claude Code — paste into CLAUDE.md at the project root

• Cursor — paste into .cursorrules at the project root

• Codex — paste into the system prompt or AGENTS.md

• Claude Desktop — paste into the system prompt field in the project settings

Pick the snippets that match your workflow. Start with "Start-of-session recall" and "Coding close" — they cover the most common case.

Start-of-session recall

Before you start, recall the project context for this task.
Then recall recent changes related to the service or component I am touching.
Search for relevant runbooks, RFCs, changelog notes, or incident notes.
Prefer `summarize_project_context` or `project_bank_view view=canonical_overview` for the first pass and then drill into `search_runbooks` or `recall_similar_incidents`.
Summarize the constraints, caveats, and likely risks before making changes.

Coding close

When the coding session ends, call `close_session` with a concise summary, service, and context.
Review the proposed `new`, `update`, `merge`, and `raw_only` actions plus the decision trace.
If the plan looks low risk, use `accept_session_changes`.
If the report is noisy or mostly exploratory, keep `save_raw_only` as the fallback.
Prefer `project_bank_view` at the next session start to confirm what became maintained knowledge.

Incident close

When incident work stabilizes, call `close_session` or `review_session_changes` with `mode=incident`.
Expect stricter review-first behavior for updates, merges, and anything touching canonical operational knowledge.
Capture impact, mitigation, rollback, and unresolved follow-ups in the summary.
Apply only the low-risk actions automatically and leave ambiguous runbook or incident changes in review.
Follow up with `recall_similar_incidents` and `project_bank_view view=incidents` if you need to compare against existing knowledge.

Migration close

When a migration session ends, call `close_session` with `mode=migration`, affected service, and the migration summary.
Prefer explicit notes about prerequisites, sequencing, rollback, and post-deploy verification.
Treat runbook replacements, caveat changes, and supersede proposals as review-first even when the textual match looks strong.
Use `accept_session_changes` only after checking the report for stale or superseded knowledge.
Finish by checking `project_bank_view view=migrations` to see the maintained migration notes.

Raw-only fallback

If the session was exploratory, ambiguous, or too noisy, skip consolidation and save only the raw summary.
Use `close_session` / `review_session_changes` to inspect the plan first, then pick `save_raw_only`.
In CLI mode, `agent-memory-mcp accept-session -raw-only ...` is the explicit override.
This keeps the raw trace without forcing weak knowledge updates into the project bank.

Before-changing-infra check

Before making infra or platform changes, recall similar fixes, migrations, incidents, and known caveats.
Search for runbooks, postmortems, changelog notes, and recent project context related to this component.
Summarize blast radius, rollback options, and operational risks before editing files.

Stewardship run

When memory has grown or a session just ended, run `steward_run` with `dry_run=true` to see what needs attention.
Review the report for duplicates, conflicts, stale entries, and canonical promotion candidates.
Check `steward_inbox` for pending review items and resolve them with `steward_inbox_resolve`.
Use `drift_scan` periodically to catch memories that reference files or docs that have changed.
Use `verification_candidates` to find knowledge that has not been verified recently.

Temporal recall

When you need to understand what was true at a specific point in time, use `recall_as_of` with an RFC3339 timestamp.
To trace how knowledge about a topic evolved over time, use `knowledge_timeline`.
When superseding an old decision or runbook, use `mark_outdated` with the superseding entry ID to build a proper chain.

HTTP mode (Docker, remote server, shared instance)

Start the server in HTTP mode:

# Standalone
MCP_HTTP_MODE=http \
MCP_HTTP_HOST=127.0.0.1 \
MCP_HTTP_PORT=18080 \
MCP_HTTP_AUTH_TOKEN=replace-with-long-random-token \
agent-memory-mcp

# Or with Docker
cd deploy/docker
docker compose --env-file .env.shared up -d --build

Then point your HTTP-capable MCP client or proxy at:

http://localhost:18080/mcp

The /mcp endpoint supports the MCP Streamable HTTP transport: JSON-RPC requests go over POST, and clients that need a server-push channel (Cursor and similar) open it with GET and Accept: text/event-stream. The server keeps that stream alive with periodic keepalive comments. A plain GET without the SSE Accept header still returns 405.

curl -N -H "Accept: text/event-stream" \
  -H "Authorization: Bearer $MCP_HTTP_AUTH_TOKEN" \
  http://localhost:18080/mcp

For retrieval inspection in a browser, open:

http://localhost:18080/console

The console is a lightweight UI for:

• running document, raw-memory, and canonical-knowledge queries

• comparing normal vs debug mode for document retrieval

• inspecting source types, trust/freshness, and score breakdowns

In shared mode, the page itself is static, but live queries from the console still require the same bearer token as /mcp.

For shared HTTP mode:

• default bare-metal bind is MCP_HTTP_HOST=127.0.0.1; this is the safe local default

• for shared/container deployments set MCP_HTTP_HOST=0.0.0.0

• set MCP_HTTP_AUTH_TOKEN to require Authorization: Bearer <token> on /mcp

• startup now fails on non-loopback binds without MCP_HTTP_AUTH_TOKEN, unless you explicitly set MCP_HTTP_INSECURE_ALLOW_UNAUTHENTICATED=true

• keep /health for load balancer or container health checks

• terminate TLS at a reverse proxy or load balancer

• do not expose the service directly on the public internet without auth and TLS

• use deploy/nginx/agent-memory-mcp.conf as the starting reverse proxy recipe

• use docs/SHARED_SERVICE.md for the full local -> team laptop -> shared service path

CLI commands

MCP tools reference

Memory tools

RAG tools

File tools

Engineering workflow tools

Stewardship tools

Temporal knowledge tools

Tool grouping mode (token efficiency)

Every MCP client loads the full JSON schema of every tool at initialize time — before your first message. With ~40 tools that schema payload alone can occupy tens of KB of the model's context window on every session. Two secondary costs compound it: LLMs get measurably worse at picking the right tool as the count climbs past ~20–40, and frequent session reloads re-pay the whole cost.

Set MCP_TOOL_GROUPING=true to collapse the core toolset into a handful of grouped meta-tools, each dispatching by a required action discriminator:

// Grouped form                          // Equivalent legacy form
{ "name": "memory",                      { "name": "store_memory",
  "arguments": {                           "arguments": { "content": "…" } }
    "action": "store", "content": "…" } }

Groups: repo · memory · memory_admin · engineering · search · session, plus the index_documents and project_bank_view singletons — the default surface drops from 41 tools to 8 (~42% smaller schema payload).

• Opt-in, zero regression. Default is false; the flag only changes what tools/list returns.

• Both call forms always work. tools/call accepts the grouped form (memory + action=store) and the legacy name (store_memory) regardless of the flag, so existing scripts never break.

• Administrative & steward tools stay individual. Rarely listed in high-volume agent runs, and steward_inbox_resolve already uses its own action argument — grouping deliberately leaves them ungrouped.

Trade-off: each grouped call carries a slightly larger per-call schema (the union of its actions' arguments). Prefer grouping for high-volume agent runs where discovery cost dominates; leave it off for interactive debugging where seeing each tool by name is clearer. Policy reference: docs/concepts/lifecycle.md covers the related archive-sweep surface.

Configuration

All configuration is via environment variables. See .env.example for the full list.

Config files are loaded in this order (each file only fills in values not already set):

1. --config /path/to/file (explicit path, skips chain)

2. .env in the current directory

3. ~/.config/agent-memory-mcp/config.env (XDG)

4. $(brew --prefix)/etc/agent-memory-mcp/config.env (Homebrew)

For solo local mode, copy .env.example to .env in your project root. For brew services, the config is auto-created at $(brew --prefix)/etc/agent-memory-mcp/config.env.

Hot-reload

When running as a service (HTTP mode), the config file is watched for changes every 30 seconds. RAG-related settings (index dirs, embedding keys, enabled/disabled) are applied without restart. HTTP settings (port, host) require a full restart.

You can also force an immediate reload:

kill -HUP $(pgrep agent-memory-mcp)

Key variables

Data paths

The server creates these directories under MCP_DATA_PATH:

• rag-index/ -- SQLite vector index for document search

• memory-store/ -- SQLite database for agent memories

The recommended solo-local preset stores them under .agent-memory/.

Indexing safety controls

RAG indexing scans supported docs and engineering text files, but you can further reduce risk with explicit controls:

• built-in excluded directories such as .git, .agent-memory, node_modules, logs, and .terraform

• MCP_INDEX_EXCLUDE_DIRS for repo-relative path excludes such as docs/private,runbooks/internal

• MCP_INDEX_EXCLUDE_GLOBS for glob-style excludes such as docs/internal/*.md

• MCP_REDACT_SECRETS=true to redact common secret-like lines and private key blocks before indexing

This is especially important if you use hosted embedding providers or shared HTTP mode.

Automatic session tracking

When the MCP server is running with the default session-tracking policy, it keeps a lightweight background session buffer.

Current behavior:

• successful MCP tool calls are grouped into an active session automatically

• idle timeout or server shutdown triggers a background close_session run

• clients can explicitly flush or checkpoint the active session with notifications/session_event and event=task_done|final_summary|checkpoint|reset

• raw session summaries are persisted automatically

• low-risk updates can auto-apply under the existing safe_auto_apply policy

• risky or ambiguous changes are stored as review inbox items instead of silently rewriting maintained knowledge

• periodic raw checkpoints provide crash-recovery breadcrumbs during long sessions

To inspect the inbox, use project_bank_view view=review_queue or agent-memory-mcp project-bank -view review_queue. To close an item after manual review, use resolve_review_item or agent-memory-mcp resolve-review-item <id>.

Example notification payload:

{
  "jsonrpc": "2.0",
  "method": "notifications/session_event",
  "params": {
    "event": "task_done",
    "summary": "Incident stabilized, workaround verified, follow-up is to replace the temporary fix.",
    "context": "payments",
    "service": "api",
    "mode": "incident",
    "tags": ["done", "verification"]
  }
}

If you want to tune or disable this behavior, use MCP_SESSION_TRACKING_ENABLED, MCP_SESSION_IDLE_TIMEOUT, MCP_SESSION_CHECKPOINT_INTERVAL, and MCP_SESSION_MIN_EVENTS.

Index integrity and recovery

Document indexing now treats chunk updates and tracking metadata as one logical state.

• each run marks the index state as dirty before changing chunks

• a successful final commit flips the state back to ready together with indexed_files, embedding_model, and last_indexed

• if a run is interrupted or the final tracking-state commit fails, the next index_documents / agent-memory-mcp index run detects the dirty state and forces a rebuild

This makes incremental indexing more predictable after crashes, provider interruptions, or storage errors.

Source-aware ingestion

The indexer now classifies engineering sources and carries that metadata into retrieval.

Supported source types:

• docs for README.md and general Markdown docs

• adr and rfc for architecture decision and RFC-style documents

• changelog for CHANGELOG.md and release-note style docs

• runbook and postmortem for operational knowledge

• ci_config for GitHub Actions, GitLab CI, and Jenkins pipeline files

• helm, terraform, and k8s for source-aware infra files

Use source_type when you want to narrow retrieval to a specific class of knowledge:

agent-memory-mcp search -source-type runbook "ingress rollback"
agent-memory-mcp search -source-type adr "cache invalidation decision"

The MCP semantic_search tool also accepts source_type and debug.

Hybrid retrieval

Search now uses multiple ranking signals instead of cosine similarity alone.

Current ranking signals:

• semantic similarity from the active embedding model

• keyword/BM25-like scoring across chunk title, path, and content

• source_type filtering when you want a narrower retrieval set

• recency boost for recently updated operational context

• source-aware weighting so runbooks, changelogs, ADRs, and other source classes can rank higher for matching query intent

The retrieval pipeline now works in two stages:

• semantic top-K candidate generation from the vector index

• keyword top-K candidate generation from a precomputed in-memory keyword index

Only the merged candidate set is reranked. This keeps shared-service retrieval more predictable as the indexed corpus grows.

This means a strong keyword hit in a runbook or changelog can outrank a semantically similar but less task-relevant document.

Trust-aware retrieval

Retrieval now carries explicit trust metadata for both stored memories and indexed docs.

Each result can expose:

• source_type

• confidence

• last_verified_at

• owner

• freshness_score

What this means in practice:

• accepted decisions and verified runbooks rank above draft or low-confidence notes when the textual match is similar

• ADRs, runbooks, postmortems, and changelogs can carry different trust weights even before you add a full canonical layer

• CLI search / recall and MCP semantic_search / recall_memory now show trust summaries in human-readable output

Engineering workflow tools also stamp stored entries with last_verified_at so fresh operational knowledge is easier to trust and rank.

Explainable retrieval

Use debug mode when you want retrieval to explain why a document was returned.

CLI:

agent-memory-mcp search -source-type runbook -debug "ingress rollback"

MCP:

• call semantic_search with debug: true

• keep debug unset or false for the normal compact response

Debug mode adds:

• applied filters such as source_type=runbook

• ranking signals used for the response

• candidate counts: indexed, filtered out, discarded as noise, returned

• per-result trust summary: source, confidence, freshness, owner, verified

• per-result score breakdown: semantic, keyword_raw, keyword_normalized, recency_boost, source_boost, confidence_boost, final_score

• per-result applied boosts, for example keyword_match or source_type:runbook

Retrieval console

If you want a faster inspection workflow than raw CLI or JSON-RPC calls, use the built-in console in HTTP mode:

/console

What it is good for:

• compare normal vs debug retrieval side by side

• inspect document search, raw memory recall, and canonical knowledge recall in one place

• see source type, trust layer, confidence, freshness, owner, and verification time

• open raw JSON for the same structured response the UI is rendering

The console is intentionally lightweight and does not replace MCP tools or CLI workflows.

Engineering workflow tools

These MCP tools map domain-specific workflows onto the existing memory and retrieval backends.

Recommended starting points:

• store_decision for architectural or operational choices such as disabling HPA or pinning an ingress version

• store_incident for short-lived operational facts you want to recall during active debugging

• store_runbook for procedural steps, rollback instructions, and verification notes

• store_postmortem for durable incident learnings and action items

• close_session when you want an explicit end-of-session plan with rationale, traces, and review-safe actions

• accept_session_changes when the close-session report is low risk and you want to persist the raw summary plus apply safe updates

• resolve_review_item when the background inbox already contains a reviewed item and you want to clear it without deleting the audit trail

• search_runbooks when you need a fix path and want both memory-stored runbooks and indexed runbook docs

• recall_similar_incidents when you are triaging an outage or regression

• summarize_project_context at session start to get a compact operational briefing

• project_bank_view when you want maintained knowledge by view instead of raw recall results, including review_queue for pending background decisions

These workflow tools also add verification metadata so that retrieval can treat newly stored operational knowledge as fresher and more trustworthy than anonymous raw notes.

Memory consolidation

The memory layer now supports a manual consolidation workflow without deleting historical notes.

Use these MCP tools when the same project knowledge starts to drift:

• merge_duplicates to consolidate repeated notes into one primary memory and archive the rest as merged duplicates

• mark_outdated to demote stale runbooks, superseded decisions, or obsolete incident notes without losing them

• promote_to_canonical to mark the current best memory as canonical knowledge

• conflicts_report to surface duplicate_candidates, status_conflict, and multiple_canonical groups

Current behavior:

• merged or archived memories stay accessible, but trust-aware recall pushes them down

• canonical memories get a trust boost and higher minimum importance

• outdated or superseded memories are automatically archived and downranked

• conflict reporting is manual and safe: nothing is deleted unless you explicitly choose to delete it

Zero-ops task-memory consolidation

Every closed task leaves behind working memories (Task started, per-phase notes, Session close, auto-extracted review items). Left alone they accumulate linearly and keep surfacing in recall for tasks that are already done. The service consolidates them automatically — no cron, no manual cleanup, no config:

• A background loop (on by default, MCP_ARCHIVE_SWEEP_INTERVAL, default 1h) sweeps archived tasks: durable entries (procedural, or importance ≥ 0.70) are promoted, the rest are marked outdated. It runs a first pass shortly after startup, which also backfills any archive that accumulated before the loop existed.

• Zero configuration. With MCP_TASK_ARCHIVE_ROOTS unset it watches the <MCP_ROOT>/tasks/archive convention; a missing directory is a silent no-op.

• Safe by construction. Promotion goes through the provenance gate: a conversational-origin memory is routed to the review queue, never auto-canonicalized (see Memory poisoning defense).

• The end_task and sweep_archive tools default to the same consolidating behavior, so an explicit /end-task consolidates immediately.

Policy details (state derivation, promotion threshold, idempotency, symlink/traversal guards): docs/concepts/lifecycle.md. Turn the loop off with MCP_ARCHIVE_SWEEP_ENABLED=false.

Canonical knowledge layer

The project now exposes two distinct layers:

• raw memory: captured notes, incidents, decisions, and procedural memories as they were stored

• canonical knowledge: confirmed entries projected from memories promoted with promote_to_canonical

What this changes:

• list_canonical_knowledge gives you the current confirmed knowledge set without raw noise

• recall_canonical_knowledge searches only canonical entries

• summarize_project_context surfaces canonical knowledge before raw memory sections when canonical entries exist

• trust summaries now show layer=raw, layer=canonical, or layer=document

Migration story:

• existing memories do not need a schema migration

• promote any high-value existing memory with promote_to_canonical

• legacy workflow memories that only have tags like decision or service:api are still recognized by the canonical layer

Knowledge stewardship

The stewardship layer provides automated and manual knowledge maintenance.

steward_run executes a full maintenance cycle in one call:

• scans for duplicate candidates (memories with matching entity/service/context/subject)

• detects conflicting entries (multiple canonical, status disagreements)

• flags stale entries (not verified within the configured threshold)

• suggests canonical promotion candidates (high-importance, active, recognized engineering type)

• generates a structured report with per-action rationale

• with dry_run=false, applies safe actions and sends the rest to the stewardship inbox

drift_scan compares memory entries against live repo files:

• detects source_changed when a referenced file was modified after the memory was last verified

• detects source_missing when a referenced file path no longer exists

• flags stale_unverified when entries exceed the stale threshold

verification_candidates ranks memories that need verification:

• never-verified canonical entries: high urgency

• entries with verification_failed or needs_update status: high urgency

• stale entries beyond threshold: medium urgency

steward_inbox is the single place for all review-required actions. Resolve items with steward_inbox_resolve using actions like merge, mark_outdated, promote, verify, suppress, or defer.

Stewardship is auto-enabled in HTTP mode when memory is available. Configure thresholds and mode via MCP_STEWARD_* environment variables. See steward_policy for runtime configuration.

For a detailed guide, see Stewardship Guide.

Temporal knowledge

Memories can carry temporal metadata that tracks when knowledge was valid and how it evolved:

• valid_from / valid_until — the time window during which this knowledge was true

• superseded_by / replaces — bidirectional links forming supersession chains

• observed_at — when knowledge was first observed (may differ from created_at)

recall_as_of retrieves knowledge that was valid at a specific timestamp. This is useful for questions like "what was our database strategy in January?" or "what changed between these two dates?"

knowledge_timeline shows the chronological evolution of entries matching a query, ordered by valid_from.

When mark_outdated is called with a superseding entry, the system automatically sets valid_until on the old entry and valid_from + replaces on the new entry, building a navigable chain.

Security And Operations

Core deployment guidance:

• solo local: keep MCP_HTTP_MODE=stdio, prefer MCP_EMBEDDING_MODE=local-only if you need no-send semantics

• shared HTTP mode: set MCP_HTTP_HOST=0.0.0.0, set MCP_HTTP_AUTH_TOKEN, keep TLS at the reverse proxy, and scope MCP_ALLOW_DIRS narrowly

• indexing: exclude private runbooks or credentials docs with MCP_INDEX_EXCLUDE_DIRS / MCP_INDEX_EXCLUDE_GLOBS

• backup: either copy .agent-memory/ or use agent-memory-mcp export for memory-only backups

Reference docs:

• Stewardship Guide

• Security Policy

• Threat Model

• Backup And Restore

• Shared Service Guide

Architecture

┌──────────────────────────────────────────────────┐
│              MCP Protocol Layer                   │
│            (stdio or HTTP/JSON-RPC)               │
├────────────┬──────────┬───────────┬───────────────┤
│Memory Tools│RAG Tools │File Tools │Steward Tools  │
├────────────┼──────────┼───────────┼───────────────┤
│MemoryStore │RAGEngine │ PathGuard │ Steward       │
│  (SQLite)  │          │           │  Service      │
│            │┌────────┐│           │  Scheduler    │
│ Embedder◄──┤│DocSvc  ││           │  Inbox        │
│            ││VecSvc  ││           │  Drift/Verify │
│            │└────────┘│           │  Policy       │
│            │ (SQLite)  │           │  (SQLite)     │
└────────────┴──────────┴───────────┴───────────────┘

Embedding providers

The server supports three embedding providers in auto mode:

1. Jina AI (primary) -- jina-embeddings-v3, native 1024 dimensions, multilingual

2. OpenAI (fallback) -- text-embedding-3-small or any OpenAI-compatible API. Native dimension is 1536; the server requests dimensions=1024 via the OpenAI MRL parameter to match the rest of the stack

3. Ollama (local fallback) -- bge-m3, native 1024 dimensions, runs locally for free

All three are normalized to the same vector dimension (MCP_EMBEDDING_DIMENSION, default 1024), but they are not interchangeable: each model has its own embedding space. Matching dimensions do not make cosine similarity safe across different models — that is why the server tags every memory with its embedding_model and refuses to mix them at recall time.

What auto mode means in practice:

• new memories or new RAG chunks use the first provider that is currently available

• existing memories keep the embedding_model they were created with

• semantic recall skips memories whose embedding_model does not match the current query model and falls back to text matching for those records

• RAG search refuses to query an index built with a different model and asks you to rebuild it

This avoids the dangerous case where provider fallback returns confident but incorrect semantic matches.

If you change provider or model intentionally, treat it as a migration:

# rebuild document index for the new embedding model
agent-memory-mcp index

# re-embed stored memories for the new embedding model
agent-memory-mcp reembed

# inspect how many memories still belong to older models
agent-memory-mcp stats -json

You can increase the dimension via MCP_EMBEDDING_DIMENSION for higher accuracy (for example 3072 with text-embedding-3-large), but any dimension or model change requires re-indexing and re-embedding.

If you set MCP_EMBEDDING_MODE=local-only, hosted providers are skipped entirely and only Ollama is used for embeddings.

Why this is better for users

• you can keep using auto mode without silent corruption of semantic recall

• you can intentionally migrate to another embedding model without losing stored knowledge

• you can audit model usage with agent-memory-mcp stats

• you can prefer local-only operation without worrying about fallback to hosted providers

macOS service installation

Homebrew (recommended)

brew install ipiton/tap/agent-memory-mcp
brew services start agent-memory-mcp

See Installation Options for details and config location.

Manual (legacy)

./scripts/install-macos.sh

This builds the binary, creates a .env file, and installs a launchd service that auto-starts on login.

Manual control:

# Status
launchctl list | grep com.agent-memory-mcp

# Start
launchctl load ~/Library/LaunchAgents/com.agent-memory-mcp.plist

# Stop
launchctl unload ~/Library/LaunchAgents/com.agent-memory-mcp.plist

Troubleshooting / FAQ

"no embedding provider available" or "embedder not configured"

The server tries Jina → OpenAI → Ollama in auto mode and stops at the first available one. If none is reachable, embeddings (and therefore semantic recall) are disabled.

• set at least one of JINA_API_KEY, OPENAI_API_KEY, or run Ollama with bge-m3 pulled (ollama pull bge-m3)

• in MCP_EMBEDDING_MODE=local-only only Ollama is consulted; verify OLLAMA_BASE_URL (default http://localhost:11434) responds

CLI agent-memory-mcp store … without an embedder still saves the memory but skips the vector — the record will only be retrievable via text/keyword search until you run agent-memory-mcp reembed.

"embedding model mismatch — index built with X, current is Y"

The RAG index and stored memories are tagged with the embedding model that produced them. If you switch provider or model, retrieval refuses to mix vector spaces:

agent-memory-mcp index            # rebuild RAG index for the new model
agent-memory-mcp reembed          # re-embed stored memories
agent-memory-mcp stats -json      # check how many memories still belong to older models

"address already in use" — port 18080 taken

Another agent-memory-mcp (or unrelated service) holds the port:

lsof -nP -iTCP:18080 -sTCP:LISTEN
brew services list | grep agent-memory-mcp

If two instances were started, stop one (brew services stop agent-memory-mcp or kill <pid>) or change MCP_HTTP_PORT.

brew services start agent-memory-mcp does not start the daemon

Most common causes:

• the formula was installed as a Cask before — uninstall first: brew uninstall --cask agent-memory-mcp && brew install ipiton/tap/agent-memory-mcp

• the config file is malformed: cat $(brew --prefix)/etc/agent-memory-mcp/config.env

• log: tail -f $(brew --prefix)/var/log/agent-memory-mcp/mcp-diagnostics.log

Hooks not firing in Claude Code

After agent-memory-mcp setup, restart Claude Code (hooks load at process start). Verify the merge:

agent-memory-mcp setup --dry-run     # show what would be written
jq '.hooks' ~/.claude/settings.json  # inspect actual config

If you upgraded via brew upgrade, re-run agent-memory-mcp setup --force so the hook command points to the new binary path. See docs/HOOKS.md.

"bind to 0.0.0.0 without auth is not allowed"

In HTTP mode the server refuses non-loopback binds without MCP_HTTP_AUTH_TOKEN. Either:

• set the bearer token (MCP_HTTP_AUTH_TOKEN=$(openssl rand -hex 32)) — recommended

• or, for tightly controlled environments only, set MCP_HTTP_INSECURE_ALLOW_UNAUTHENTICATED=true

Memory database keeps growing

• run agent-memory-mcp project-bank canonical_overview and … review_queue to spot raw session summaries that should be promoted or marked outdated

• enable stewardship: MCP_STEWARD_MODE=scheduled triggers periodic dedup/stale scans

• prune session checkpoints with the dedup window (MCP_CHECKPOINT_DEDUP_THRESHOLD, MCP_CHECKPOINT_DEDUP_WINDOW)

• back up first: agent-memory-mcp export > backup.json (see docs/BACKUP_RESTORE.md)

recall_multihop returns empty

The multi-hop graph walks the (subj, rel, obj) triple corpus. It's empty until you either:

• enable MCP_TRIPLE_EXTRACTOR_ENABLED=true and write new memories (extraction is async on every store)

• backfill existing memories with agent-memory-mcp index-triples (idempotent, supports --resume)

Both paths require MCP_TRIPLE_EXTRACTOR_* (see Key variables) — extraction calls an OpenAI-compatible /chat/completions endpoint.

"config not reloading" in HTTP/service mode

Changes are picked up within ~30s. Force an immediate reload:

kill -HUP $(pgrep agent-memory-mcp)

Note: HTTP host/port require a full restart (brew services restart agent-memory-mcp). Only RAG-related settings hot-reload.

Development

# Build
make build

# Run
make run

# Test
make test

# Full quality gate (lint + tests + smoke)
make quality-gates

For repo layout, code style, and PR conventions see docs/CONTRIBUTING.md.

License

MIT