How to use the agentmemory MCP Tools Map agent skill

What is an agent skill and what does this one do?

An agent skill is a folder with a SKILL.md file that gives an AI agent extra instructions it can read when a task matches the skill's description. This one is a map of agentmemory's MCP tools: it groups them into families such as capture, retrieval, sessions, knowledge graph, structured slots, and governance and health checks.

The skill does not store or fetch memories itself. The separately running agentmemory server does that. The skill only helps the agent choose the right tool and call it correctly.

Install the skill in Claude Code or the Claude apps

Download the ZIP for this skill from this page and unzip it. For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/agentmemory-mcp-tools/
• Project: .claude/skills/agentmemory-mcp-tools/
• Claude apps: Settings → Capabilities → Skills

Connect agentmemory before you use the skill

The skill only describes tools that must already be connected. You need a running agentmemory server or daemon and an MCP connection to it. For Pi and OMP, the source says to connect with --with-mcp to enable the broader workflows.

If you use a native memory provider, the agent may see fewer tools under different names. In that case, the agent should rely on the live tool inventory rather than this skill's list.

• A running agentmemory server/daemon is required.
• An MCP connection to agentmemory is required (Pi and OMP: start with --with-mcp).
• REFERENCE.md and TROUBLESHOOTING.md must stay in the skill folder.

How to trigger the skill with natural requests

The skill's description says to use it when choosing which memory tool to call, when a tool name or argument is unclear, or when answering what agentmemory can do via MCP. Ask in those terms. The skill is marked user-invocable: false in its front matter, so it is meant to be picked up by the agent from context rather than run as a slash command.

Requests that match the skill include asking which memory tool fits a task, asking what parameters a tool takes, or asking what agentmemory can do through MCP.

• Which agentmemory tool should I use to save a design decision?
• What arguments does memory_recall take?
• What can agentmemory do over MCP?

Example: save a lesson and recall it later

The quick start in the skill describes a two-step pattern. First, call memory_save with content (the insight), concepts (comma-separated keywords), and files (comma-separated paths). Later, call memory_smart_search with query, limit, and targetLayer set to "memory" to retrieve saved knowledge.

Omit targetLayer if you also want to search captured observations. According to the skill, memory_smart_search runs hybrid BM25, vector, and graph-expanded search.

• Save: memory_save with content, concepts, files
• Recall: memory_smart_search with query, limit, targetLayer: "memory"

Choose the right tool and avoid argument mistakes

The skill recommends picking the narrowest tool for the task. Use memory_smart_search for open recall, memory_recall when you already have a focused query, and memory_sessions for session listings. Check exact parameter names and which are required in REFERENCE.md before calling.

Pass only documented fields, because REST handlers whitelist fields and drop unknown ones. For budgeted memory_recall results, tell apart matched_count: 0 from excluded_by_budget > 0. If results are clipped or omitted, expand the IDs with memory_smart_search and expandIds, keeping the same scope filters.

• Open recall: memory_smart_search
• Focused query: memory_recall
• Session listings: memory_sessions
• matched_count: 0 means no match; excluded_by_budget > 0 means matches were cut by the budget.

Safety notes: scripts and network

According to the skill's metadata, this skill does not run scripts and does not use the network. It is a text reference that the agent reads. Any real memory activity happens through the agentmemory server you run and connect to.

Because the skill mentions REST equivalents, see the separate agentmemory-rest-api skill for HTTP details. Keep in mind that the tools themselves can save, search, and delete memory, so review what your agent saves and which governance tools it calls.

• Runs scripts: no
• Uses network: no
• Governance tools include memory_governance_delete.

Troubleshoot "MCP tool not available"

If the agent reports that an MCP tool is unavailable, the cause may be a stopped MCP process, a disconnected daemon, or a reduced tool surface. The bundled TROUBLESHOOTING.md walks through the checks in order, starting with identifying your host.

In Claude Code, the troubleshooting file suggests inspecting /plugin list and /mcp. After configuration changes, restart or reload the affected host connection.

• Claude Code: inspect /plugin list and /mcp.
• Pi or OMP: inspect /agentmemory-status and the MCP inventory.
• Restart or reload the host connection after configuration changes.