pathmark
Pathmark is a local MCP memory server for saving, retrieving, and curating shared intent and evidence across coding agents.
Save raw searchable evidence with
remember, and propose durable approval-gated conclusions withcreate_conclusion.Search, recall, and retrieve context via
search_memory,recall_memory,get_context, andask_memory/chat(approved conclusions first).Review and manage conclusions with
list_conclusions,list_pending_conclusions,approve_conclusion, andreject_conclusion.Inspect usage and provenance with
session_trace,rate_recall,audit_memory, anddoctor_memory.Maintain the local store with
delete_memory,update_memory,supersede_memory,purge_memory,consolidate_memory, andcompact_memory.Back up, export, and sync scopes with
backup_memoryandexport_memory.Generate approved USER/PROJECT/AGENT snapshots with
get_memory_snapshot, and inspect configuration withget_config.
Allows the server to use Ollama (or any OpenAI-compatible local gateway) as a synthesis provider for ask_memory, via the openai-compatible mode.
Allows the server to use OpenAI-compatible chat completion APIs for memory synthesis via the ask_memory tool, using the PATHMARK_OPENAI_BASE_URL and API key.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@pathmark remember that I prefer using TypeScript for new projects"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Pathmark
Carry intent across agents without turning stale code facts into hidden memory.
What's New — v0.1.17
Pathmark v0.1.17 stops Claude Code from gating every memory lookup:
pathmark setup claude-code --apply-permissionsadds allow rules for Pathmark's ten local read-only tools, so recall and search run without a permission prompt, and in auto mode without a safety-classifier round-trip that can fail transiently;it backs up
~/.claude/settings.json, adds only missing rules, keeps every other setting, and never rewrites a file it cannot parse;the allow list is tested against the tools' own
readOnlyHint/openWorldHintannotations, so anything that writes, deletes, or may call an external model always keeps its prompt.
See the v0.1.17 release notes or the complete changelog. The npm badge above always shows the currently published version.
Pathmark gives Codex, Claude Code, opencode, Gemini CLI, Cursor, and any MCP-capable harness one local intent and provenance layer. Save decisions, constraints, preferences, and approved conclusions once. Use them from the next agent without pasting a recap.
Code remembers implementation. Pathmark remembers intent. Repository code, architecture, tests, CI, and intentional agent instructions remain authoritative for how the software works. Raw sessions are searchable evidence, not automatically trusted truth.
Your context stays on disk at ~/.pathmark/memory/memory.jsonl. You do not need an account, hosted database, API key, or vendor backend to start.
Related MCP server: Memryzed
OpenAI Build Week 2026
Pathmark is a Developer Tools submission for OpenAI Build Week 2026. The project existed before the challenge, so the submission is deliberately scoped to the meaningful extension built after the submission period opened on July 13, 2026.
During the eligible period, Codex with GPT-5.6 helped audit and extend Pathmark from a working local memory layer into safer long-running developer infrastructure:
fixed a reproduced multi-process SQLite index race;
added revision history, superseding, expiration, retention, diagnostics, backup, compaction, and preview-first hard purge;
added namespace-scoped reads and writes plus default secret redaction;
added scoped import/export, optional AES-256-GCM portable exports, local hybrid reranking, and portable harness ingestion;
hardened CI and npm delivery with required CodeQL and dependency review, immutable Action pins, protected tags, OpenSSF analysis, and SLSA provenance.
The primary Codex session for this work is 019f5fc3-d7e6-7b41-8a30-d161c90b98fb. The qualifying release range is v0.1.6 through v0.1.7; the pre-challenge baseline is commit 4c0e87dfdbd2ba4c643abd8b887cc228bdb08b73.
See the Build Week implementation record for the before/after boundary, commit evidence, Codex collaboration details, and a fast judge test.
Why Pathmark
You do not work in one tool. You ask Codex to patch, Claude Code to review, opencode to clean up, and Gemini CLI to challenge the plan. Each tool starts cold unless you carry the context across.
Pathmark gives those tools one place to read and write intent and evidence:
One local JSONL store across harnesses.
Standard MCP tools include
remember,search_memory,recall_memory,session_trace,rate_recall,consolidate_memory,audit_memory, and conclusion-firstchat/ask_memory.Client-side synthesis by default, so your coding agent reads the context and answers.
Optional Codex CLI, local command, and OpenAI-compatible synthesis modes.
Plain files you can inspect, back up, delete, or migrate.
Pathmark stays provider-neutral. Codex gets one optional synthesis preset. The core server works with any MCP client that can use local tools.
Pathmark requires Node.js 22.5 or newer.
Cross-Harness Memory
You switch tools during a coding session:
Codex fixes the failing test.
Claude Code reviews the patch.
opencode cleans the diff.
Gemini CLI challenges the approach.
Pathmark keeps the notes in one store.
Point each harness at the same store:
Codex \
Claude Code \
opencode > Pathmark MCP > ~/.pathmark/memory/memory.jsonl
Gemini CLI /
Cursor /Install Pathmark in each harness and point them at the same PATHMARK_STORE_DIR. One tool saves raw context with remember or proposes a durable conclusion with create_conclusion; an approved conclusion and raw evidence can then be recovered with recall_memory, search_memory, get_context, or ask_memory.
Pathmark sits below the agents as an intent, evidence, and provenance bus for your coding workflow.
Tools
Pathmark exposes these MCP tools:
Tool | Purpose |
| Save raw searchable evidence. Raw evidence is not treated as durable approved intent. |
| Propose a higher-signal durable conclusion or preference. Approval is required by default before recall. |
| Search memories and conclusions. |
| Transparent recall: returns context plus the exact memory IDs, timestamps, sources, matches, tags, and previews used. Accepts optional |
| Return a bounded chronological audit trail for one session: prompts, exact injected memory IDs, redacted tool inputs/results, and answers. |
| Label exact IDs from a |
| Return compact context for a task or question. |
| List approved saved conclusions. |
| Review bounded, paginated pending conclusion proposals. |
| Atomically approve a proposal, optionally correcting text/tags and recording the reviewer. |
| Retain a rejected proposal in the audit trail while permanently excluding it from recall. |
| Generate a bounded USER/PROJECT/AGENT snapshot from approved canonical conclusions. |
| Review a bounded unsynthesized evidence batch and optionally stage evidence-backed proposals. Nothing is auto-approved. |
| Soft-delete a memory or conclusion by id. |
| Correct a record while preserving prior versions. |
| Replace an outdated record with a linked current record. |
| Preview or apply permanent deletion by id, namespace, tags, source, or date. |
| Measure capture-to-recall behavior, unused records, recall age, duplicates, stale raw hits, and whether precision labels exist. |
| Report duplicates, deleted/expired records, conclusions, and index health. |
| Preview or apply deduplication, retention, and physical cleanup with an automatic backup. |
| Create a point-in-time canonical JSONL backup. |
| Export a scoped mergeable JSONL bundle, optionally encrypted. |
| Return an approved-conclusion answer or scoped raw context, exact provenance, and a recall ID for feedback. |
| Chat-compatible alias for |
| Show local store configuration. |
Quick Start
npm install -g pathmarkThen add the MCP server to your client.
Prefer npm for normal installs. To test the current GitHub main branch directly:
npm install -g --install-links=true github:hacksurvivor/pathmarkGenerate a setup snippet for your harness:
pathmark setup list
pathmark setup claude-code
pathmark setup opencode --json
pathmark setup gemini-cli
pathmark setup kimiSee docs/compatibility.md for Codex, Claude Code, opencode, Gemini CLI, OpenClaw, Hermes Agent, Grok CLI, Kimi, GLM, and generic MCP setups.
Codex
codex mcp add pathmark -- pathmarkCodex users can also enable auto-capture:
pathmark codex install --replace-legacy-hooksWhen you want the visible "what memory did you use?" entry in Codex, Claude Code, Cursor, opencode, Gemini CLI, Grok-compatible MCP hosts, or any other MCP harness, call the recall_memory tool before answering. Codex session start injects an approved conclusion snapshot. Before non-trivial prompts, Codex recalls approved conclusions first and uses fresh scoped raw evidence only as a bounded fallback. recall_memory remains the portable visible trace across harnesses.
Claude Code
claude mcp add --scope user pathmark -- pathmark
pathmark setup claude-code --apply-permissionsThe second command lets Pathmark's local read-only tools (recall, search, diagnostics) run without a permission prompt, including in auto mode, where they would otherwise wait on the safety classifier. It backs up ~/.claude/settings.json, adds only the missing allow rules, and leaves everything else untouched. Tools that write or delete memory still ask.
--scope user makes Pathmark available in every project; Claude Code's default scope only covers the current directory. For automatic capture and session-start recall (including after context compaction), merge the hooks block from pathmark setup claude-code into ~/.claude/settings.json.
Claude Code's built-in auto-memory lives in per-project folders the other agents cannot see. Bring it into the shared store as evidence (re-runnable; edits update in place with history, and records you delete in Pathmark stay deleted):
pathmark import-native claude-code --dry-run
pathmark import-native claude-codeopencode / Gemini CLI
Use the generated snippets:
pathmark setup opencode
pathmark setup gemini-cliClaude Desktop
Add this to your Claude Desktop MCP config:
{
"mcpServers": {
"pathmark": {
"command": "pathmark",
"env": {
"PATHMARK_STORE_DIR": "~/.pathmark/memory"
}
}
}
}Cursor
Add the same command to Cursor's MCP server settings:
{
"mcpServers": {
"pathmark": {
"command": "pathmark"
}
}
}Local Development
npm install
npm test
npm run coverageRun directly:
PATHMARK_STORE_DIR=.pathmark npm run devImport Legacy Memory
Pathmark can import a compatible local JSONL memory store without deleting or moving the source files.
npm run import:legacy -- --source-dir ~/old-codex-memoryDefaults:
Legacy source: ~/.pathmark/legacy/codex
Pathmark target: ~/.pathmark/memory/memory.jsonlThe importer creates a memory.jsonl.backup-* file before writing, uses deterministic ids so reruns skip duplicates, and redacts obvious KEY=..., TOKEN=..., PASSWORD=..., and Bearer ... values.
It uses the same store lock as live MCP and Codex writers, so an import cannot overwrite records captured concurrently.
Use a dry run first when migrating another machine:
npm run import:legacy -- --source-dir ~/old-codex-memory --dry-runCodex Auto-Capture
Install Pathmark as the Codex memory adapter:
pathmark codex install --replace-legacy-hooksThis registers the Pathmark MCP server, enables Codex hooks, and removes old compatible hook commands from Codex. It does not delete or move memory files.
The Codex adapter is proactive by default:
user prompts, final assistant answers, and tool activity are captured locally; intermediate Codex commentary is excluded;
tool activity records include bounded redacted input previews and hashes, status, exit code, duration, and changed files when the hook provides them;
tool-output hashes are captured for correlation, while output text remains private by default and requires
PATHMARK_CODEX_CAPTURE_TOOL_OUTPUTS=on;activity records expire after 30 days and are physically capped at 5,000 records by default;
session start/resume generates one bounded USER/PROJECT/AGENT snapshot from approved canonical conclusions and does not inject raw session history;
each non-trivial user prompt searches approved workspace/project conclusions first, then approved global or explicitly named-project conclusions;
only when no approved conclusion matches, at most two raw records from the current workspace/project/session may be injected as a high-confidence fallback;
raw fallback records must be within the separate automatic-recall horizon, 30 days by default; the full raw archive remains available to explicit
search_memoryandrecall_memorycalls;raw cross-project history is never injected automatically; promote durable cross-project intent through the approval workflow;
broad cross-project history remains available through explicit
search_memory/recall_memorycalls without silently entering every prompt;matching memory is injected quietly by default, without adding a raw
recall_memorytool result to the conversation;set
PATHMARK_CODEX_VISIBLE_RECALL=onwhen debugging or auditing to make Codex callrecall_memorywith the exact pre-capture result IDs and workspace tag; this explicit mode omits the redundant fullrecordscopy;legacy transport envelopes and assistant progress updates are excluded from session-start and proactive relevance results, while realtime delegation envelopes retain only their current
<input>payload;records containing Pathmark boundary escapes, instruction-override patterns, or invisible Unicode controls are tagged
memory-quarantinedand excluded from automatic recall; injected previews are escaped and explicitly treated as untrusted historical data;no matching memory means no extra context is injected.
Durable extraction is approval-gated by default. create_conclusion creates a pending proposal; pending and rejected conclusions stay in the canonical JSONL audit trail but are structurally excluded from normal search, exact-ID recall, prompt injection, and snapshots. Use list_pending_conclusions, then approve_conclusion or reject_conclusion. Conclusions created before this workflow are treated as already approved for backward compatibility. Raw remember records remain searchable evidence and are not promoted automatically.
The session snapshot is generated from the same canonical store rather than maintained as a second flat file. It is frozen in the session-start hook output; prompt-time scoped recall remains dynamic.
Set PATHMARK_CODEX_PROACTIVE_RECALL=off if you want Codex hooks to capture memory but stop prompt-time recall.
Set PATHMARK_CODEX_VISIBLE_RECALL=on only when you want an explicit audit/debug recall_memory tool call. Proactive prompt-time recall remains active when this is off.
recall_memory is a point-in-time record of memory used before an answer. It intentionally does not include commands that run later. Use session_trace with the exact session ID to inspect the chronological prompt → injected memories → tools/results → final-answer trail. When explicitly enabled, output previews are capped at 2,000 characters and redacted before storage; hashes preserve correlation without storing output text by default. Upgrading an existing cursor migrates to final-answer-only parsing without duplicating previously captured user or final-answer turns. When the original transcript is available, exact legacy phase: "commentary" records are soft-deleted by timestamp and text during that migration.
Use --replace-legacy-hooks when you want Pathmark hooks to take over from earlier compatible hook commands. Without it, Pathmark installs alongside existing hook commands.
Check the adapter status:
pathmark codex statusThe status output is JSON and includes Pathmark hook state, MCP registration state, legacy hook presence, the active store paths, and the current record count.
Remove Pathmark hooks and MCP registration without deleting memory:
pathmark codex uninstallConfiguration
Variable | Default | Description |
|
| Directory for |
|
| Default search limit. |
|
| Automatically inject relevant Pathmark context before non-trivial Codex prompts. Use |
|
| Opt into an audit/debug |
|
| Store bounded redacted tool-output previews. Output hashes, status, duration, and exit codes remain available when this is off. |
|
| Generate a bounded approved-conclusion snapshot at Codex session start/resume. |
|
| Character budget for generated snapshots; clamped to 500–12000. |
|
| Prompt-time eligibility horizon for raw evidence. |
|
| Maximum fresh raw records injected when no approved conclusion matches. Clamped to 0–2. |
|
| At session start, nudge Codex to review a bounded evidence batch when scoped raw history is accumulating without conclusions. |
|
| Minimum unsynthesized user/assistant records before the proactive consolidation nudge appears. |
|
| Stage new conclusions for explicit approval. Set |
|
|
|
| unset | Command provider: receives a synthesized prompt on stdin and writes an answer on stdout. |
|
| Codex provider command. |
| unset | Optional Codex model override. |
|
| OpenAI-compatible API base URL. |
| unset | OpenAI-compatible API key. |
| unset | Model id for OpenAI-compatible synthesis. |
|
| Synthesis command timeout. |
| unset | Default namespace applied consistently to MCP reads and writes. |
|
| Redact common secret-shaped values on |
|
| Retention policy used by compaction; |
|
| Automatic lifetime for recall/tool activity records; |
|
| Physical cap for activity records; oldest activity is removed automatically. |
| unset | Optional trusted local embedding/vector or hybrid reranker. Strict kind/tag/namespace filters are applied before candidates leave the store; the command receives query/candidates as JSON on stdin and returns ranked memory ids. |
|
| Maximum candidates sent to the optional reranker. |
|
| Timeout for the optional reranker. |
| unset | Passphrase for AES-256-GCM portable exports/imports. Never returned by |
|
| Cross-process wait limit for index initialization or rebuild. |
Synthesis Modes
Pathmark separates memory from reasoning.
client
Default. The MCP server returns relevant memory context, and your MCP client model synthesizes the answer. This works across Codex, Claude Desktop, Cursor, and any other MCP client without giving Pathmark a model credential.
PATHMARK_SYNTHESIS_PROVIDER=client pathmarkcommand
Use any local subscription or model CLI that accepts a prompt on stdin and writes an answer to stdout:
PATHMARK_SYNTHESIS_PROVIDER=command \
PATHMARK_CHAT_COMMAND="your-ai-cli --model your-model" \
pathmarkThis is the general path for users with another paid subscription CLI or a local model runner.
codex
Use the proven Codex CLI bridge. It runs a controlled, non-interactive codex exec turn with hooks and memories disabled to avoid recursion:
PATHMARK_SYNTHESIS_PROVIDER=codex \
PATHMARK_CODEX_MODEL=gpt-5.5 \
pathmarkThis is useful for Codex users who have persisted ChatGPT/Codex CLI auth locally but do not want to add an OpenAI API key. Pathmark sends the synthesis prompt through stdin, runs Codex in an empty temporary workspace, ignores project rules, and exposes only a minimal environment. Memory records are treated as untrusted data rather than executable instructions.
openai-compatible
Use any provider that exposes /chat/completions, including many Kimi, GLM/Z.ai, OpenRouter, LiteLLM, Ollama-compatible gateways, and self-hosted routers:
PATHMARK_SYNTHESIS_PROVIDER=openai-compatible \
PATHMARK_OPENAI_BASE_URL=https://api.provider.example/v1 \
PATHMARK_OPENAI_API_KEY=... \
PATHMARK_OPENAI_MODEL=... \
pathmarkThis mode affects MCP ask_memory / chat and CLI pathmark chat. Regular MCP tools still store and retrieve local memory without a model provider.
Setup CLI
pathmark setup <client> prints copy-paste setup for common harnesses. Add --json when you want structured output for scripts.
Supported targets:
codex
claude-code
claude-desktop
cursor
opencode
gemini-cli
generic
openai-compatible
commandAliases include claude, gemini, kimi, glm, and z-ai.
Gemini CLI setup includes portable SessionStart, BeforeAgent, AfterTool, and AfterAgent hooks for automatic scoped recall and capture. Other harnesses can feed exported transcripts through the generic ingestion surface:
pathmark ingest --client=claude-code --namespace=my-project < transcript.json
pathmark ingest --client=opencode --namespace=my-project < transcript.jsonMemory chat, consolidation, maintenance, and portable sync
Maintenance commands preview destructive changes unless --apply is present:
pathmark chat "What did we decide about release signing?" --namespace=my-project
pathmark consolidate --namespace=my-project
pathmark consolidate --namespace=my-project --cursor=LAST_RECORD_ID
PATHMARK_SYNTHESIS_PROVIDER=codex pathmark consolidate --namespace=my-project --apply
pathmark feedback --recall-id=RECALL_ID --relevant=MEMORY_ID --irrelevant=OTHER_ID
pathmark audit --days=30
pathmark audit --namespace=my-project --days=90
pathmark doctor
pathmark compact
pathmark compact --apply --retention-days=90
pathmark purge --namespace=old-client
pathmark purge --namespace=old-client --applypathmark chat and the MCP chat / ask_memory tools search approved conclusions first. Multi-intent questions can return separate conclusions for separate clauses. In default client mode, approved conclusions produce a safe extractive answer; a configured codex, command, or openai-compatible provider can synthesize richer prose. Raw fallback requires an explicit scope (--namespace / tags) or kind: memory, preventing unscoped cross-workspace history from entering chat.
Every matched chat query records recall activity and returns a recallId when the store is writable. Abstentions and read-only stores return recallId: null. Use MCP rate_recall or pathmark feedback with exact recalled IDs to label relevance. pathmark audit reports precision.status: "labeled", measured precision, and label coverage once feedback exists.
pathmark consolidate is preview-first. With default client synthesis it returns the bounded evidence and exact instructions for the host agent, which can call create_conclusion with supporting evidenceIds. When more eligible evidence remains, the result includes nextCursor and remainingAfterBatch; pass the cursor to review the next stable page. With a configured server-side synthesis provider, it previews structured candidates; --apply stages them as pending conclusions for approve_conclusion or reject_conclusion. It never auto-approves extracted intent.
Applied compaction and purge create a backup before replacing the canonical file. Soft deletion remains available through delete_memory; hard purge physically removes selected records from JSONL and rebuilds the derived index.
pathmark audit is read-only. It separates all raw records from consolidation-eligible user/assistant evidence, then reports capture-to-recall ratio, actionable synthesis backlog, recall age, exact duplicates, stale raw hits, and scope/missing-reference signals. Precision remains unlabeled until explicit feedback exists; Pathmark never substitutes a heuristic for a user label.
Use scoped exports and merge imports as the transport-neutral sync layer:
pathmark export --namespace=my-project --output=project.jsonl
pathmark import project.jsonl --namespace=my-projectFor an encrypted portable bundle, configure the passphrase outside the command line:
PATHMARK_EXPORT_KEY='use-a-secret-manager' pathmark export --encrypted --output=project.pathmark
PATHMARK_EXPORT_KEY='use-a-secret-manager' pathmark import project.pathmarkPathmark does not silently upload these files. Move them through a trusted filesystem, backup tool, or sync provider of your choice.
Optional hybrid retrieval
Default retrieval stays local SQLite FTS. To enable semantic or embedding-backed reranking without forcing a model dependency, set PATHMARK_RERANK_COMMAND to a trusted local command. It receives one JSON object on stdin containing query and candidates, and must return a JSON array of ranked record ids (or { "ids": [...] }). If it fails or times out, Pathmark falls back to lexical results.
Data Format
Pathmark stores newline-delimited JSON at:
~/.pathmark/memory/memory.jsonlmemory.jsonl remains the canonical source of truth. Pathmark also maintains a derived, disposable search index at memory.index.v5.sqlite. Index filenames are schema-versioned so old and new MCP processes can coexist during a rolling restart. The index is rebuilt automatically when the JSONL file changes outside Pathmark, and inactive index versions can be deleted safely after their processes stop.
Each record is inspectable:
{
"id": "uuid",
"kind": "memory",
"text": "The user prefers local-first tools.",
"tags": ["preference"],
"source": "mcp",
"createdAt": "2026-06-29T00:00:00.000Z",
"updatedAt": "2026-06-29T00:00:00.000Z"
}Deletes are soft deletes by default: the record gets a deletedAt timestamp. Use preview-first purge_memory or pathmark purge --apply for physical erasure. Updates preserve up to 50 prior versions, superseded records link to their replacement, and expired records are excluded from recall. The raw automatic-recall horizon is separate from storage retention: evidence can age out of proactive injection while remaining explicitly searchable.
Malformed JSONL lines are skipped rather than crashing every tool. pathmark codex status reports their count as invalidRecordCount so the source file can be repaired deliberately.
Roadmap
Provider presets for common local AI CLIs where stable commands exist.
Encrypted store option.
Hosted sync as an opt-in layer, not a requirement.
Native auto-capture packages for additional harness plugin systems beyond Codex and Gemini CLI.
Example recipes for Codex, Claude Desktop, Cursor, ChatGPT, and local LLM tools.
Positioning
Pathmark gives your agents a shared working memory that stays on your machine.
Switch agents. Keep the context.
Bring your own subscription. Keep your memory local.
Author and citation
Pathmark is created and maintained by Sergey Moloman, an AI product engineer.
Machine-readable authorship and citation metadata are available in CITATION.cff and codemeta.json.
License
MIT
Available Tools
25 toolsapprove_conclusionApprove conclusionAIdempotent
Approve one pending conclusion, optionally correcting its text or tags. The transition is atomic and auditable.
Inputs: id (required), note, tags, text, +2 more.
ask_memoryAsk memoryARead-onlyIdempotent
Ask approved conclusions first, then scoped or explicitly requested raw evidence. Returns an answer, exact provenance, and a recallId for feedback.
Inputs: question (required), kind, tags, limit, +1 more.
audit_memoryAudit memory valueARead-onlyIdempotent
Measure capture-to-recall behavior, unused records, recall age, duplicate rate, stale raw hits, and available precision evidence without changing memory.
Inputs: days, tags, namespace.
backup_memoryBack up memory storeC
Create a point-in-time copy of the canonical local JSONL store.
Inputs: destination.
chatChatBRead-onlyIdempotent
Chat with Pathmark using approved conclusions first and only scoped or explicitly requested raw fallback. Returns an answer, provenance, and recallId.
Inputs: question (required), kind, tags, limit, +1 more.
compact_memoryCompact memory storeADestructive
Preview or apply exact deduplication, expired-record removal, retention, and deleted-record purging. Applied runs create a backup.
Inputs: dedupe, confirm, dropDeleted, retentionDays.
consolidate_memoryConsolidate raw evidenceBIdempotent
Prepare a bounded unsynthesized evidence batch and, when server synthesis is configured, preview or stage evidence-backed conclusion proposals. Proposals are never auto-approved.
Inputs: days, tags, apply, cursor, +3 more.
create_conclusionCreate conclusionC
Propose a durable higher-signal conclusion. Approval is required by default before it can be recalled.
Inputs: text (required), tags, source, expiresAt, +2 more.
delete_memoryDelete memoryADestructiveIdempotent
Soft-delete a saved memory or conclusion by id.
Inputs: id (required).
doctor_memoryDiagnose memory storeARead-onlyIdempotent
Report duplicate, deleted, expired, conclusion, invalid-record, and index health counts without changing data.
Inputs: no named parameters.
export_memoryExport memoryB
Export a scoped, mergeable JSONL bundle for another Pathmark installation or trusted sync transport.
Inputs: destination (required), kind, tags, encrypted, +2 more.
get_configGet Pathmark configurationARead-onlyIdempotent
Show the local Pathmark Memory store location and enabled optional features.
Inputs: no named parameters.
get_contextGet contextCRead-onlyIdempotent
Return compact local memory context for a task or question.
Inputs: kind, tags, limit, query, +1 more.
get_memory_snapshotGet approved memory snapshotBRead-onlyIdempotent
Generate a bounded USER/PROJECT/AGENT snapshot from approved canonical conclusions only.
Inputs: tags, charLimit, namespace.
list_conclusionsList conclusionsCRead-onlyIdempotent
List saved durable conclusions.
Inputs: tags, limit, namespace.
list_pending_conclusionsList pending conclusionsARead-onlyIdempotent
List bounded approval-gated conclusion proposals. Pending records are never returned by normal memory search.
Inputs: tags, limit, offset, namespace.
purge_memoryHard purge memoryCDestructive
Preview or permanently remove matching records from the canonical store. A backup is created before an applied purge.
Inputs: id, tags, before, source, +2 more.
rate_recallRate recalled memoriesC
Attach explicit relevance labels to one exact Pathmark recall so audit_memory can report measured precision.
Inputs: recallId (required), note, relevantIds, irrelevantIds.
recall_memoryRecall memoryBRead-onlyIdempotent
Transparent recall for any MCP-capable harness. Use this at task start or before answering to show exactly which memories were used.
Inputs: ids, kind, tags, limit, +3 more.
reject_conclusionReject conclusionADestructiveIdempotent
Reject one pending conclusion while retaining it in the canonical audit trail and excluding it from recall.
Inputs: id (required), note, decidedBy.
rememberSave raw evidenceAIdempotent
Save raw searchable evidence. Durable intent should use the approval-gated conclusion workflow.
Inputs: text (required), tags, source, expiresAt, +1 more.
search_memorySearch memoryBRead-onlyIdempotent
Search saved local memories and conclusions.
Inputs: kind, tags, limit, query, +1 more.
session_traceSession traceARead-onlyIdempotent
Show a bounded chronological audit trail for one captured session: prompts, exact injected memory IDs, redacted tool inputs/results, and answers.
Inputs: sessionId (required), limit, includeOutputs.
supersede_memorySupersede memoryC
Replace an outdated memory with a linked current record while preserving history.
Inputs: id (required), text (required), kind, tags, +3 more.
update_memoryUpdate memoryC
Correct an existing memory while preserving its prior versions in local history.
Inputs: id (required), tags, text, source, +2 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 25 tools
chat and ask_memory are nearly identical in purpose—both answer using approved conclusions first with raw fallback and return provenance plus a recallId—creating clear ambiguity. Additionally, search_memory, get_context, recall_memory, and ask_memory overlap in retrieval behavior, though their descriptions partially clarify output differences. The raw-evidence tools (remember, create_conclusion, consolidate_memory) are more distinct but still require careful reading.
Most tools follow a consistent verb_noun snake_case pattern such as create_conclusion, approve_conclusion, delete_memory, and export_memory. A few bare-verb or noun-like names like remember, chat, and session_trace deviate slightly, but the overall convention is recognizable and predictable.
At 25 tools, this sits at the heavy end of the borderline range and feels like more surface than most agents will need. The count is defensible for a full memory lifecycle system covering capture, recall, conclusions, audit, and maintenance, but it risks overwhelming users.
The tool set covers the core memory lifecycle well: capture raw evidence, propose and approve conclusions, search and recall, update/delete/supersede, audit, compact, backup, and export. Minor gaps exist—there is no import tool to complement export, and no direct get-by-id retrieval—but agents can work around these via search and export/backup workflows.
Maintenance
Related MCP Connectors
Shared AI memory. ChatGPT, Claude, Cursor and any MCP app read and write the same memory.
Persistent memory for AI agents across Claude, ChatGPT and any MCP client.
Persistent AI memory shared across Claude, ChatGPT, coding agents, and compatible MCP clients.
Shared memory MCP server for Claude Code, Codex, Gemini CLI and Cursor. Cited, editable.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA shared, local-first memory layer for AI CLIs, providing persistent, layered memory across Claude Code, Gemini CLI, and other MCP-aware clients.1MIT
- FlicenseNot gradedqualityCmaintenanceLocal-first cross-agent memory for AI coding agents. Persistent, shared memory over MCP — what you tell one agent can be recalled by another — with all data stored in a single local SQLite file, no cloud and no API keys.-
- AlicenseAqualityCmaintenanceLocal-first persistent memory for coding agents and MCP clients. It keeps important project context across sessions and reduces wasted tokens by retrieving only relevant memories instead of replaying unnecessary history.8MIT
- AlicenseBqualityBmaintenanceProvides local-first durable memory and session continuity for AI coding agents over MCP, enabling context across restarts without cloud services or telemetry.1MIT