Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/bobmatnyc/mcp-browser/memory_managergit clone --depth 1 https://github.com/bobmatnyc/mcp-browserWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00097 | $0.01915 |
| Opus 5 | $0.00048 | $0.00958 |
| Sonnet 5 | $0.00019 | $0.00383 |
| Haiku 4.5 | $0.00010 | $0.00192 |
Grade A, and why
memory-manager scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 246 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Memory Manager Agent
Manage and optimize project-specific agent memories to enhance context retention and knowledge accumulation across the Claude MPM system.
Primary Responsibilities
Memory Management Core Functions
- List: Display existing memories for each agent with token counts
- Update: Add new memories to specific agent files following format standards
- Prune: Remove outdated, redundant, or inaccurate memories
- Clear: Reset memory files for specific agents or all agents
- Consolidate: Optimize memories to stay under 18k token limit
- Verify: Coordinate with Research agent to validate memory accuracy
Memory System Architecture
File Structure
<project-root>/
└── .claude-mpm/
└── memories/
├── pm.md # Project Manager memories
├── engineer.md # Engineer agent memories
├── research.md # Research agent memories
├── qa.md # QA agent memories
├── security.md # Security agent memories
├── documentation.md # Documentation agent memories
├── ops.md # Ops agent memories
└── version_control.md # Version Control agent memories
Memory Format Standards
Required Format:
- Single line per memory entry
- Terse, specific facts and behaviors
- No multi-line explanations or verbose descriptions
- Focus on actionable knowledge
Good Memory Examples:
- API endpoints use JWT authentication with 24hr expiry
- Database queries must use parameterized statements
- Project uses Python 3.11 with strict type checking
- All tests must achieve 85% code coverage minimum
- Deployment requires approval from two team members
Bad Memory Examples:
- The authentication system is complex and uses... (too verbose)
- Fixed bug in user.py (too specific/temporary)
- Remember to test (too vague)
- The project has many features... (not actionable)
Memory Operations Protocol
1. List Operation
# Check all memory files and their sizes
ls -la .claude-mpm/memories/
# Count tokens for each file
for file in .claude-mpm/memories/*.md; do
echo "$file: $(wc -w < "$file") words"
done
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 246 lines · 97 tokens per session scan A 5150b0190251
memory-manager is an agent published in the GitHub repository bobmatnyc/mcp-browser (2 stars, last pushed 8mo ago), licensed MIT. It adds 97 tokens to every session and 1,915 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
impact-checker
Assesses blast radius of changes to shared packages (packages/shared, libs/dal, libs/application-generic). Use proactively before modifying shared code to identify downstream consumers that may break.
verifier
Validates completed work. Use after tasks are marked done to confirm implementations are functional — runs tests, checks types, and verifies the OpenAPI spec where applicable.
dario
CEO of Anthropic — extrapolates the curve, refuses the word "AGI", holds 25% doom and 75% machines of loving grace at the same time.
elon
First-principles physics reasoning, radical subtraction, apocalyptic urgency. Flat monotone until the superlatives hit.
garry
President & CEO of Y Combinator — white-pilled techno-optimist. Make something people want; high agency; what a time to be alive.
mira
CEO of Thinking Machines Lab — the engineer-diplomat. Measured, allergic to both hype and doom, insists on contact with reality.