Independent use will determine whether MCP Memory’s combination of an OKF-derived memory approach and SQLite FTS5 provides useful, fast persistent memory for agents.

state: expiredheat: lowuncertainty: highknownscott: mediumagent-memory mcp sqliteMCP Memoryfellowgeek

What is this?

The evidence title presents MCP Memory, attributed to fellowgeek, as an open-source persistent-memory implementation for AI agents that combines an approach described as derived from Google’s OKF with SQLite FTS5 retrieval. Related search snippets support the broader design pattern: local SQLite-backed agent memory can provide durable, searchable context, verbatim session replay, auditability, and low-dependency deployment. However, the supplied web results do not clearly identify the specific MCP Memory project or independently establish its OKF connection, performance, or usefulness, so those claims remain to be validated through independent use.

Why it matters to Scott

Scott already holds the core position that agent continuity belongs in durable external state and that retrieval substrates should be chosen and evaluated through real agent walks; see “RAG/Wiki Substrate Rule,” “Search What Churns,” and “Reflexive Agent Design.” MCP Memory adds no validated result yet, but independent testing could directly inform the lexical-retrieval and MCP architecture of his active MCP IP Wiki, especially whether SQLite FTS5 is sufficient as recall beneath a maintained graph.

ip:framework.rag-wiki-substrate-ruleip:concept.search-what-churnsip:framework.reflexive-agent-designdev:project.mcp-ip-wikiradar:concept.agent-memoryradar:concept.mcpradar:anansi-open-memory-apiradar:deepmem-hybrid-agent-memory

queries asked of Scott's wikis

SQLite FTS5 versus vector search for agent memory
MCP memory-server architecture and interoperability
agent-maintained wikis versus structured memory stores
local-first zero-dependency agent infrastructure
persistent session replay and agent audit trails
benchmarks for useful agent memory retrieval

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (3) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn	Show HN: MCP Memory – Fast Agent Memory Using Google's OKF and SQLite FTS5	pcbmaker20	62	35
🟧 echo.github ⭐	MCP Memory is an open-source persistent agent-memory implementation combining an approach attributed to Google’s OKF with SQLite FTS5 retrie	fellowgeek	—	—
🟧 hn	Memento – Shared, durable memory for multiple AI agents, with MCP access	rcarmo	2	0

Interpretation history

2026-08-16T21:31:14Z
No independent use, benchmark, or architectural result emerged during the observation window, so MCP Memory’s specific OKF/FTS5 value remains unvalidated and no longer merits active tracking. The broader durable MCP memory pattern remains covered by adjacent cases.
2026-08-14T20:37:47Z
The refreshed comments remain repetitive amplification of the existing files-and-grep objection, without independent use, benchmarks, or architectural findings. The adjacent MCP memory pattern is corroborated, but MCP Memory’s specific OKF/FTS5 value remains untested.
2026-08-13T20:33:10Z
The refreshed discussion remains repetitive skepticism about whether MCP/OKF adds value over files and grep; it adds no independent test of MCP Memory’s retrieval quality, latency, or practical agent outcomes. The broader durable-memory pattern is corroborated, but this implementation’s central claim is not.
2026-08-13T18:49:31Z
Memento shows that shared durable memory over MCP is an independently implemented pattern with at least one author reporting months of use, moving the case beyond a lone repository. It still does not validate MCP Memory’s OKF/FTS5 retrieval quality, while the discussion sharpens the unresolved comparison with plain files and grep.
2026-08-13T15:23:50Z
evidence attached: hn.story.49286431 — First-party artifact for shared durable multi-agent memory over MCP directly bears on the open persistent-memory hypothesis.
2026-08-13T14:43:32Z
The added engagement is amplification rather than validation; no independent testing, implementation experience, or retrieval-quality evidence changes the case’s meaning.
2026-08-13T14:37:30Z
grounded: known/medium — Scott already holds the core position that agent continuity belongs in durable external state and that retrieval substrates should be chosen and evaluated throu
2026-08-13T14:34:57Z
case created — The repository is a concrete new memory-layer artifact, but there is not yet independent evidence of retrieval quality or practical adoption.

Decision trace

08-17 07:31expireNo independent use, benchmark, or architectural result emerged during the observation window, so MCP Memory’s specific OKF/FTS5 value remains unvalidated and no longer merits active tracking. The broa
08-17 07:31alert_silentThe staleness trigger adds no consequential evidence or decision-relevant change; any future independent evaluation can reopen the case.
08-17 07:31alert_routeThe staleness trigger adds no consequential evidence or decision-relevant change; any future independent evaluation can reopen the case.
08-15 06:37repriceThe refreshed comments remain repetitive amplification of the existing files-and-grep objection, without independent use, benchmarks, or architectural findings. The adjacent MCP memory pattern is corr
08-15 06:37alert_silentEngagement rose, but the new delta contains no validated result that would affect Scott’s memory or retrieval design decisions before the next briefing.
08-15 06:37alert_routeEngagement rose, but the new delta contains no validated result that would affect Scott’s memory or retrieval design decisions before the next briefing.
08-14 06:33repriceThe refreshed discussion remains repetitive skepticism about whether MCP/OKF adds value over files and grep; it adds no independent test of MCP Memory’s retrieval quality, latency, or practical agent
08-14 06:33alert_silentNo new benchmark, usage report, or architectural finding changes Scott’s MCP or SQLite FTS5 decisions; refreshed comments can wait for the next briefing.
08-14 06:33alert_routeNo new benchmark, usage report, or architectural finding changes Scott’s MCP or SQLite FTS5 decisions; refreshed comments can wait for the next briefing.
08-14 05:21sensor_dirtycomment_update
08-14 04:49repriceMemento shows that shared durable memory over MCP is an independently implemented pattern with at least one author reporting months of use, moving the case beyond a lone repository. It still does not
08-14 04:49alert_silentThe adjacent implementation and skeptical comparisons are useful context, but no independent benchmark, target-project usage report, or architectural result changes Scott’s design decisions before the
08-14 04:49alert_routeThe adjacent implementation and skeptical comparisons are useful context, but no independent benchmark, target-project usage report, or architectural result changes Scott’s design decisions before the
08-14 04:21sensor_dirtycomment_update
08-14 03:21sensor_dirtycomment_update
08-14 02:21sensor_dirtycomment_update
08-14 01:26alert_silentMemento is a relevant alternative architecture for shared durable memory over MCP, but the only new evidence is a low-engagement project link and its author’s claim of several months’ use. There are n
08-14 01:26surface_candidateMemento is a relevant alternative architecture for shared durable memory over MCP, but the only new evidence is a low-engagement project link and its author’s claim of several months’ use. There are n
08-14 01:26alert_routeMemento is a relevant alternative architecture for shared durable memory over MCP, but the only new evidence is a low-engagement project link and its author’s claim of several months’ use. There are n
08-14 01:23attachFirst-party artifact for shared durable multi-agent memory over MCP directly bears on the open persistent-memory hypothesis.
08-14 01:23propose_attachFirst-party artifact for shared durable multi-agent memory over MCP directly bears on the open persistent-memory hypothesis.
08-14 01:21sensor_dirtycomment_update
08-14 00:43repriceThe added engagement is amplification rather than validation; no independent testing, implementation experience, or retrieval-quality evidence changes the case’s meaning.
08-14 00:43alert_silentOnly engagement increased, with no substantive new evidence about recall quality, speed, or adoption; it can wait for independent use or benchmark results.
08-14 00:43alert_routeOnly engagement increased, with no substantive new evidence about recall quality, speed, or adoption; it can wait for independent use or benchmark results.
08-14 00:38alert_silentA relevant open-source implementation appears to exist, but the only new delta is its launch and architectural claim; there are no independent tests, benchmarks, adoption signals, or concrete findings
08-14 00:38surface_candidateA relevant open-source implementation appears to exist, but the only new delta is its launch and architectural claim; there are no independent tests, benchmarks, adoption signals, or concrete findings
08-14 00:38alert_routeA relevant open-source implementation appears to exist, but the only new delta is its launch and architectural claim; there are no independent tests, benchmarks, adoption signals, or concrete findings
08-14 00:37groundScott already holds the core position that agent continuity belongs in durable external state and that retrieval substrates should be chosen and evaluated through real agent walks; see “RAG/Wiki Subst
08-14 00:34createThe repository is a concrete new memory-layer artifact, but there is not yet independent evidence of retrieval quality or practical adoption.