Log inSubmit server
PMB AI logo

PMB AI

Local-first MCP memory for coding agents, storing project context, decisions, lessons, and documents in SQLite.

InstallRepositoryWebsite
Full README Compare alternatives Embed widget Claim this listing

Install PMB AI

Pick your client, then add it in one click or paste the config.

STDIONode.js
~/Library/Application Support/Claude/claude_desktop_config.json
claude_desktop_config.json
{
  "mcpServers": {
    "pmb-ai": {
      "command": "npx",
      "args": [
        "-y",
        "pmb-ai"
      ]
    }
  }
}

Settings → Developer → Edit Config, merge this into mcpServers, then fully quit and reopen Claude Desktop.

We couldn’t confirm it starts correctly

We ran the install command but it didn't respond within our test window — this can mean a slow first-time install rather than a real problem.

npx -y pmb-ai

No response to initialize.

This is an experimental automated check and can have false negatives — missing environment variables, a slow cold install, etc. It doesn’t necessarily mean something’s wrong. Last checked Today.

Overview

The PMB AI MCP server gives MCP-aware coding agents persistent local memory backed by SQLite. It captures and retrieves project context, decisions, lessons, goals, activity, personal facts, project structure, and PDF content through MCP, with search indexes stored locally. Install it with Python, run its setup command to connect an agent, and use the CLI for indexing, recall, statistics, dashboards, and backups. Reach for it when agents need continuity across sessions without cloud storage, API keys, or an LLM call on the read path.

Use cases

  • Recall decisions and lessons from earlier coding sessions
  • Index project structure and commit intent for agent context
  • Store changing personal facts with historical values
  • Search local PDFs and research documents
  • Create and verify local SQLite memory snapshots

Key features

  • SQLite-backed persistent memory
  • MCP context preparation with lessons and project activity
  • Local project and PDF indexing
  • Automatic memory capture for supported agent workflows
  • Local dashboard with graph, timeline, and recall views
  • Verified snapshot creation and restoration

Security & health

What it can reach and how reliably it runs — from our own checks, not the maintainer’s claims.

ActiveLast checked Today

Recent health check succeeded.

8/8 checks healthy over the last 40d

Supply-chain signal

No high-severity advisories surfaced by our automated scan.

Critical 0High 0Medium 0Low 0

Scanned 12d ago via OSV.dev · pmb-ai (npm)

Adoption & maintenance

Factual signals from GitHub, npm, and our automated checks — not a rating.

GitHub stars284Stargazers on the source repository.
npm downloads28Package downloads in the last 30 days.
Last commit1mo agoMost recent push to the default branch.
Install checkInconclusiveDidn't respond in our test window — often a slow first install.
Directory activity1 viewsConfig copies, upvotes, and views on AllMCPs.

How PMB AI works

What PMB AI MCP server does

The PMB AI MCP server provides a local memory workspace for coding agents. SQLite is the durable source of truth, while local search indexes can be rebuilt beside it. Memory can include decisions, lessons, completed tasks, goals, recent activity, personal facts, project structure, commit intent, and PDF content.

The system is content-agnostic: text that an agent may need later can be stored and retrieved. Personal facts support changing values, with older values retained as history. Project indexing is aware of symbols, imports, and .gitignore rules. PDF indexing accepts a file or a directory, including recursive directory processing.

The PMB AI MCP server is designed for continuity between sessions, model upgrades, and agent changes. It also provides a local dashboard for inspecting the stored memory as an entity graph and timeline. The dashboard includes views for entities, narrative arcs, lessons, duplicates, performance, and recall debugging.

How it works

An MCP call named prepare(message) gathers memory relevant to a request. Its response can include project context, matching procedural lessons, recent activity, open goals, and active narrative arcs. Lessons include a surface_id, allowing the agent to confirm that a rule was followed later.

Memory can be written as agents work, while additional data can be added explicitly with commands such as record_keyed_fact, project indexing, PDF indexing, and tracking for modules or changes. The read path does not require an LLM call. The README reports warm recall at roughly 35 milliseconds and writes below one millisecond, though actual results depend on the environment.

The workspace remains on the local machine. The dashboard listens on 127.0.0.1:8765, and the project states that no cloud account, API key, or telemetry is required.

Setup and configuration

Install the Python package with pip install pmb-ai, then run pmb setup to detect an agent and add the MCP connection. The package provides the pmb command and also a pmb-ai alias. An npm entry point is available through npx pmb-ai setup; it installs the Python package first and then runs setup.

After setup, restart the agent. pmb warmup preloads the model for faster first recall, pmb stats reports stored data, pmb recall "query" searches from the terminal, and pmb doctor checks whether the setup is wired correctly. No environment variables or API keys are listed as required.

Tools and capabilities

  • Prepare context for a message with relevant project information, lessons, activity, goals, and arcs.
  • Store keyed personal facts while preserving earlier values.
  • Index source projects and PDFs, including recursive PDF directory indexing.
  • Track module purpose and the reasons behind new commits.
  • Inspect memory through a local dashboard with graph, timeline, lesson, duplicate, performance, and recall views.
  • Create, list, verify, and restore SQLite snapshots through the CLI.
  • Export stored data to Markdown or JSON.

The PMB AI MCP server supports Claude Code, Cursor, Codex, Windsurf, Zed, VS Code, and other MCP-aware agents named by the project. The README specifically describes automatic memory injection through hooks for Claude Code and Codex.

Limitations and notes

The project is local-first: its documented storage and dashboard behavior are local, and it does not provide cloud synchronization. Restoring a snapshot requires PMB servers, agent connections, and dashboards to be stopped first. Restore validates the snapshot and saves a verified copy of the previous state before changing memory.

The dashboard binds only to 127.0.0.1. Search indexes are rebuildable, but SQLite remains the authoritative store.

The PMB AI MCP server is distributed under Apache 2.0. Its setup and command examples assume a Python package installation, with npm available as an alternative setup path.

Read the full README View source on GitHub

Reviews

No reviews yet — be the first to share how it worked for you.

Frequently Asked Questions about PMB AI

PMB AI is an MCP server that connects MCP-aware coding agents to a local SQLite memory workspace. Its main tools and commands prepare relevant context, record keyed facts, index projects and PDFs, track module and commit information, recall stored memory, and inspect or back up the workspace without cloud storage or API keys.

Install it with `pip install pmb-ai`, then run `pmb setup` to detect an agent and configure its MCP entry. Restart the agent afterward; `pmb warmup` can preload the model for faster first recall. The npm path is `npx pmb-ai setup`.

Yes, PMB AI lists Cursor among the MCP-aware agents it can connect to. Run `pmb setup` to detect an agent and wire the MCP entry, then restart Cursor so the connection becomes available.

It stores durable memory locally in SQLite, with rebuildable search indexes kept locally as well.

Use `pmb snapshot create --note "..."` to create a snapshot, `pmb snapshot list` to view snapshots, and `pmb snapshot verify <snapshot-id>` to check one. Restore with `pmb snapshot restore <snapshot-id>` after stopping PMB servers, agent connections, and dashboards.

Related MCP servers

All in Knowledge & Memory All alternatives
Add a badge to your READMEShow it’s listed — live SVG, Markdown or HTML.

AllMCPs Directory Badge

Full Badge Customizer

Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.

Badge Style:
Live Dynamic SVG PreviewPMB AI AllMCPs Directory Badge
Markdown (GitHub README)
[![AllMCPs](https://allmcps.com/api/badge/pmb-ai?style=directory)](https://allmcps.com/mcp/pmb-ai)
HTML Embed
<a href="https://allmcps.com/mcp/pmb-ai"><img src="https://allmcps.com/api/badge/pmb-ai?style=directory" alt="PMB AI on AllMCPs" /></a>