GrayMatter
Persistent agent memory as a single Go binary: knowledge graph, TUI observability, zero infra.
Connector by angelnicolasc · GitHub account verified by the MCP Registry
Install
Pick your client. Read the repository before you give any tool access to your data.
No install instructions from the source yet. Check the repository for setup steps.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this server can do and what it asks for, from its tool list and install config. Checked Oct 25, 2026.
- Tools
- 3 read · 2 write · 0 delete · 2 not yet classified
- Secrets
- None asked for
- Install scripts
- Not scanned yet
Tools by what they do 7
Read from the source code; the running server may list different tools.
Read 3 · Look things up without changing anything.
checkpoint_resumememory_searchmemory_search_batch
Write 2 · Create, change, send or run things.
checkpoint_savememory_add
Not yet classified 2 · We could not tell from the name and description.
memory_aliasmemory_reflect
Tools (7)
Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].
checkpoint_resumeRead an agent's most recent checkpoint without modifying anything: returns its ID, creation time (RFC3339), and the saved state as indented JSON, plus a message-turn count when messages were captured. Use at session start to detect and resume interrupted work; checkpoints are created with checkpoint_save. With no checkpoint, on_missing="error" (default) returns the historical not-found error; on_missing="empty" returns a successful {"found": false, "agent_id"} result. Errors stay prose-only for storage and daemon failures.
checkpoint_savePersist a snapshot of an agent's working state so a later session can pick the task back up with checkpoint_resume. Use before major refactors or when ending a session mid-task. Each save keeps its own ID and nothing is overwritten; checkpoint_resume returns the latest one. state must be a valid JSON object and is rejected otherwise. Returns a confirmation containing the new checkpoint ID.
memory_addStore one durable fact for an agent so future sessions can recall it. Use for conclusions worth keeping — decisions, preferences, gotchas — one atomic sentence per call, not transcripts or conversation logs. To correct or retire an existing fact use memory_reflect instead. Nothing is hard-deleted: untouched facts decay on a 30-day half-life unless pinned. Returns a confirmation naming the agent.
memory_aliasDeclare that a term and another term mean the same thing in one agent's memory vocabulary, so later memory_search queries bridging the two reach the right facts. Use it when a search missed because your wording and the store's wording differ — the weak-match feedback block names this tool for exactly that case. Alias facts are never returned as memories and can be corrected with memory_reflect like any fact. Returns a confirmation naming the term and its equivalents.
memory_reflectCurate an agent's memory: add a fact, update (supersede) an existing one, forget, pin or unpin against decay, or link two knowledge-graph nodes. Supply agent_id or deprecated alias agent: at least one is required at runtime, every supplied identity must be a non-empty string containing non-whitespace characters, and agent_id wins when both are valid. Use it mid-task when you notice a contradiction, finish a task, or learn a durable preference; for a brand-new fact memory_add is simpler. update requires the exact old fact text in target; forget, pin and unpin accept the fact via target or text,…
memory_searchSearch one agent's stored facts and return the most relevant ones ranked by hybrid recall (semantic + keyword + recency). Call once with your agent_id, then again with agent_id "__shared__" for project-wide conventions. Returns a numbered list with a count header, or a "No memories found" notice when nothing matches; superseded and decayed facts never appear. Results may carry a weak-match note naming the store's nearby vocabulary: reformulate once with those terms, and declare a lasting synonym with memory_alias instead of retrying synonyms blindly. To store new facts use memory_add; to corre…
memory_search_batchAnswer SEVERAL questions against one agent's memory in a single call, run concurrently. Use it whenever you hold more than one open question; use memory_search when you have exactly one. Put EVERY open question into ONE call: queries run in parallel server-side and splitting them costs a model round trip each. A query may come back with a weak-match note naming nearby vocabulary: reformulate those once with the suggested terms; lasting synonyms belong in memory_alias. Returns `merged`, a deduplicated best-first block ready to read where a fact answering several queries appears once, plus `per_…
Signals
- Stars
- 487
- Forks
- 41
- Open issues
- 5
- Last commit
- Oct 9, 2026 (19 days ago)
- Last release
- Oct 9, 2026 (19 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- Registry name
io.github.angelnicolasc/graymatter- Version
- 0.16.4
- License
- MIT
- Repository
- github.com/angelnicolasc/graymatter
- First seen
- Oct 24, 2026
- Last checked
- Oct 25, 2026
Cite this page
GrayMatter MCP server. Agent Mart. https://agentmart.dev/connectors/angelnicolasc-graymatter (last updated 2026-10-25)