q-skillvault
OfficialThis MCP server exposes SkillVault as a local-first knowledge, memory, and workflow backend for AI agents.
Save, search, retrieve, archive, and tag prompts, skills, decisions, sessions, references, routing rules, and other vault entries.
Store long AI outputs/specs/reports as file-backed artifacts and save AI results as entries.
Compile priority/token-aware context packs in modes like profile, project, planning, session_recall, and full_brief; also return structured context bundles.
Manage projects, session wraps, handoffs, and pending/decision/learning summaries.
Create and traverse entry relations such as references, depends_on, supersedes, part_of, and related_to with graph traversal.
Compare entries via diff, list/restore entry versions, and compose series or render workflows.
Run workflows, route scenarios to workflows/skills, and inspect workflow run analytics.
Add/resolve/search/list URL bookmarks and instruction recipes with placeholder interpolation.
Save/search/get/timeline episodic memories and build memory-based project context packs.
Report vault statistics and workflow-run stats.
Integrates with GitHub issues through the sibling line factory to support issue-to-PR workflows, including planning, building, review, waiting for CI, and repair on CI failure.
q-skillvault · Qu@ntum
Local-first knowledge operating system for developers and AI agents.
Store, search, and retrieve prompts, skills, workflows, decisions, project memory, session summaries, and long AI outputs from the SkillVault CLI in the broader local-first tool suite.
_____ _ _ _ __ __ _ _
/ ____| (_) | | \ \ / /__ _ __ | |__ ___| |
| (___ | | | | | \ \ / / _ \ '_ \| '_ \ / _ \ |
\___ \| | | | | \ V / __/ | | | |_) | __/ |
____) |_|_| | | \_/ \___|_| |_|_.__/ \___|_|
|_____/ |_|Author: Gabriel Magallon Sanchez - Qu@antum
Repository: https://github.com/QuantumEdu/q-skillvault
Codename: Qu@ntum
Status: v3 — Workflow bridge + LifeOS taxonomy + workflow analytics + entry versioning + skill pack export + issue-to-PR factory (line)
Versioning: Semantic Versioning MAJOR.MINOR.PATCH starting at v3.0.0 — MINOR bumps per merged feature PR, PATCH per merged fix PR, MAJOR only on breaking changes. Single source of truth: internal/version/version.go.
Binary size: ~7 MB
Dependencies: Zero frameworks. Only modernc.org/sqlite.
Language: Go 1.26+
License: MIT
Docs
Guide | Description |
Install, init, and first 6 steps in 5 minutes | |
Full CLI reference — all commands with flags | |
Variable detection, frontmatter, and injection guide | |
MCP server setup for Claude Code / OpenCode | |
Real-world workflow: project → skills → context → session | |
Clean Architecture deep-dive, data flows, design decisions | |
| |
Full-text index invariant, drift detection, and why FTS5 | |
Memory write path: dedupe, supersession, lifecycle hints and their guarantees |
Related MCP server: ellmos-homebase-mcp
What SkillVault Does
SkillVault is a local knowledge and workflow layer for humans and AI agents:
Store reusable prompts, skills, references, decisions, session summaries, and workflow notes.
Classify entries by type and by LifeOS-aligned purpose (
WORK,KNOWLEDGE,LEARNING,RELATIONSHIP,STATE,OBSERVABILITY).Import workflow-builder YAML into SkillVault workflows and phase-skill entries.
Route natural scenarios to the right workflow or skill with
skillvault route <scenario>.Run workflows from the CLI or via MCP with structured JSON-RPC-compatible output.
Serve agent tools over MCP (
run_workflow,route_scenario, search, context, graph, artifacts, and more).Run the sibling
linefactory for one GitHub issue (plan→build→review→wait_ci→ready, withrepairon CI failure). Seedocs/line.md.
Pain Problem
Working with AI coding agents creates repeated friction:
Prompts and skills scatter across chats, files, tools, and GitHub repos.
Agents get overloaded if every skill is installed globally.
Long AI outputs (PDF analyses, specs, reports) are valuable but pollute context if saved into prompts.
Project decisions are forgotten across sessions.
Workflows miss steps when there's no reusable checklist.
Context is either too little or too much — never the right amount.
SkillVault solves this by becoming a local source of truth for everything your agents need, with retrieval designed for both humans and AI.
Innovation: The Qu@ntum Context Layer
Most vaults just store things. SkillVault Qu@ntum delivers the right context when agents need it.
7 context modes — profile, project, workflow, skill, planning, session_recall, full_brief
Priority-based compilation — user feedback > project state > decisions > workflows > sessions > artifact summaries > references
Token-aware truncation — respects
max_chars, drops lowest-priority sections firstOutput: clean, structured text ready for agent injection:
# CONTEXT PACK
## Scope
Project: MyApp
Mode: planning
## User Preferences
- Prefer practical architecture
- Use TDD for core domain behavior
## Active Decisions
- SQLite for local storage
- No cloud sync in v1
## Suggested Next Action
Generate implementation plan from spec.Installation
# Prerequisites: Go 1.26+
git clone --recurse-submodules https://github.com/QuantumEdu/q-skillvault.git
cd q-skillvault
# Build and install core binary
make install
# Or install the full suite (skillvault + line factory + telemetry + q-secrets)
make install-all
# Initialize the vault
skillvault init
# Verify
skillvault versionThat's it. Single binary with zero CGO dependencies. No external database servers or heavy frameworks.
Self-update: Use
skillvault updateto rebuild and reinstall from source. Configure the source repo withSKILLVAULT_REPOenv var and target path withSKILLVAULT_INSTALL_PATH. Seeskillvault update --helpfor details.
Quickstart
# Check your local setup first
skillvault doctor
# Create a project
skillvault add-project --name "MyApp" --description "My application"
# Save a reusable skill
skillvault add-entry \
--title "Clean Architecture Review" \
--type skill \
--purpose KNOWLEDGE \
--summary "Checklist for reviewing clean architecture compliance" \
--project myapp \
--tags "architecture,review"
# Search your vault
skillvault find "architecture"
# Filter by LifeOS-style purpose
skillvault search "architecture" --purpose KNOWLEDGE
# Capture a per-project pending item without breaking flow
skillvault pending add --project myapp "Update presentation"
# Review or resolve it later
skillvault pending review --project myapp
skillvault pending list --project myapp --query presentation
skillvault pending show update-presentation
skillvault pending done update-presentation
# Save a long AI output as an artifact (stored on disk, indexed in DB)
skillvault save-artifact \
--title "PDF Analysis - Security Audit" \
--type pdf_analysis \
--content "$(cat /tmp/long-analysis.md)" \
--project myapp \
--tags "security,audit"
# Get compact context for your agent
skillvault context --mode planning --project myapp --max-chars 5000
# Optional: build and open the lightweight TUI dashboard
make build-tui
./skillvault-tui tui
# Write a dated backup snapshot
skillvault backup
# Wrap up a session with decisions
skillvault session-wrap \
--project myapp \
--summary "Reviewed auth middleware" \
--decisions "Use JWT,not sessions" \
--pending "Add refresh token rotation"Workflow bridge quick path
# Import workflow-builder YAML into SkillVault
skillvault import-workflow --file .agent/skills/research/workflow.yaml --project myapp
# Add a routing entry that maps scenarios to a workflow or skill
skillvault add-entry \
--title "Research route" \
--type routing \
--purpose WORK \
--summary "Route research scenarios" \
--body $'research:\n workflow: research-workflow' \
--tags workflow-route
# Resolve what should handle a scenario
skillvault route research
skillvault route research --json
# Execute a workflow pipeline from the CLI
skillvault run research-workflow input.md --save output.mdArchitecture
Core Principle
DB decides. Disk remembers. Qu@ntum delivers.DB decides: what exists, what type, status, relations, how to find it (SQLite + FTS5).
Disk remembers: long artifacts, AI outputs, specs, reports stored as files in
objects/YYYY/MM/.Qu@ntum delivers: compact, filtered, priority-sorted context for agents.
Package Map
cmd/skillvault/
├── internal/cli/ # 25+ CLI commands (stdlib, no Cobra)
├── internal/mcp/ # 24 MCP tools over stdio JSON-RPC 2.0
├── internal/api/ # HTTP API (local only)
├── internal/app/ # Use cases: save, search, context, session, refs, memory, pipeline
├── internal/domain/ # Pure entities + validators
├── internal/db/ # SQLite + FTS5 stores + embedded migrations
├── internal/files/ # Artifact filesystem (objects/YYYY/MM/)
├── internal/context/ # Qu@ntum context compiler (7 modes)
├── internal/security/ # Secret scanner (4 regex patterns)
├── internal/vars/ # Variable detection + injection + frontmatter parser
├── internal/export/ # Import/Export with conflict resolution
└── internal/api/ # HTTP REST serverStorage Layout
~/.skillvault/
├── vault.db # SQLite + FTS5 (metadata, search, relations)
├── objects/ # Long artifact files
│ └── YYYY/MM/
│ └── <slug>.<ext> # Content-addressed by SHA256
├── exports/ # JSON exports
└── cache/ # Temporary cacheRule: Content goes to DB when small and frequently retrieved. Content goes to filesystem when long, or a final document, or an AI output worth preserving.
CLI Commands
Command | Description | Example |
| Create vault directories + DB |
|
| Check whether the local vault is ready |
|
| Save a reusable entry |
|
| FTS5 search with filters |
|
| Get entry by ID or slug |
|
| Save a long file-backed artifact |
|
| Save an AI result as a vault entry |
|
| Print a ready-to-paste MCP client snippet |
|
| Compile Qu@ntum context pack |
|
| Create a project |
|
| List all projects |
|
| Capture, list, and resolve per-project pending items |
|
| Save, search, and copy URL bookmarks |
|
| Save and interpolate command recipes with placeholders |
|
| Archive an entry |
|
| Create a workflow (JSON file) |
|
| Import workflow-builder YAML |
|
| Render workflow as checklist |
|
| Resolve scenario → workflow or skill |
|
| Execute a workflow pipeline step by step |
|
| Save session with decisions |
|
| Visualize entry graph |
|
| Manage graph edges (alias) |
|
| Manage graph edges |
|
| Show version history for an entry |
|
| Restore an entry to a previous version |
|
| Sovereign episodic memory CLI with FTS5 and deduplication |
|
| Index pi-memory.md files |
|
| Write a dated backup under |
|
| Two-way bidirectional sync & merge with GitHub backup repo |
|
| Export vault to JSON or skill pack ( |
|
| Import vault from JSON or skill pack |
|
| Show vault version |
|
| Vector similarity between two entries |
|
| Show vault statistics and entry counts |
|
| Rebuild and reinstall binary from source |
|
| Run q-secrets (optional encrypted secret manager) |
|
q-secrets integration (optional)
SkillVault can run q-secrets as a subprogram for local encrypted secret management.
What is q-secrets?
q-secrets is a local encrypted secret manager that stores secrets in a SQLite database with keyring-backed master key encryption. It supports init, add, get, run (inject secrets into a child process), list, delete, rotate, export, and import commands.
Installation
q-secrets is included as a git submodule in the q-skillvault repository. After cloning q-skillvault, initialize the submodule:
# Clone q-skillvault with submodules
git clone --recurse-submodules https://github.com/QuantumEdu/q-skillvault.git
# Or, if you already cloned without --recurse-submodules:
git submodule update --init --recursive
# Install both skillvault and q-secrets
make install-all
# Or install via q-skillvault (recommended)
skillvault init --with-secrets
# Or install q-secrets separately
make install-q-secrets
# OR
skillvault secrets installUsage
# Initialize q-secrets (creates ~/.q-secrets/secrets.db and generates keys)
skillvault secrets init
# Add secrets to a profile
skillvault secrets add pi KEY
# Or pipe a value from a secure source without placing it in argv
secret-source | skillvault secrets add pi GITHUB_TOKEN --stdin
# Run with a minimal inherited environment
skillvault secrets run pi --clean-env -- opencode
# List profiles (names only)
skillvault secrets list
# Retrieve a secret
skillvault secrets get pi KEY
# Run a command with secrets as environment variables
skillvault secrets run pi -- opencode
# All q-secrets commands are available
skillvault secrets version
skillvault secrets rotate pi
skillvault secrets export piOn Unix, skillvault secrets replaces the wrapper process so q-secrets receives
signals directly and preserves the child exit status. On Windows it retains the
subprocess wrapper.
Binary resolution order
The secrets command resolves the q-secrets binary by checking, in order:
Q_SECRETS_BINenvironment variable (full path override)Same directory as the
skillvaultexecutableq-secrets/q-secretsrelative to the q-skillvault repository root (dev mode)q-secretsinPATH
If the binary is not found, run make install-q-secrets or set Q_SECRETS_BIN.
Updates
skillvault update will automatically rebuild and reinstall q-secrets if the
q-secrets/ directory exists in the q-skillvault repository. Set SKIP_Q_SECRETS=1 to
skip the q-secrets rebuild.
Agent Telemetry System
Local-first observability for CLI LLM coding agents. Collects, redacts, and persists tool calls, token usage, quality signals, and agent lifecycle events to SQLite.
Components
telemetryd— Unix socket daemon that validates, redacts, hashes, and stores eventstelemetryctl— CLI to query runs, events, and daemon statustelemetrywrap— CLI wrapper that infers telemetry events from any commandOpenCode plugin — Native
EventEmitteremitting 9 canonical event types
Quick Start
# Build the binaries
go build -o ~/tools/telemetryd ./cmd/telemetryd/
go build -o ~/tools/telemetryctl ./cmd/telemetryctl/
# Start the daemon
telemetryd &
# Check daemon status
telemetryctl status
# List runs
telemetryctl run list
telemetryctl run recentSecurity
SHA-256 arg hashing with 32-byte random salt (
~/.telemetry/salt, 0600 perms)Built-in regex redaction: OpenAI keys, Bearer tokens, auth headers,
--api-keyflagsBase64 entropy scanning →
scanned-warningflagZero plaintext secrets in logs, DB, or CLI output
For full documentation, see docs/telemetry.md.
MCP Tools (39 default, 44 engram-compat)
For AI agents (Claude Code, OpenCode, Antigravity, Pi, Codex):
Tool | Description |
| Save a prompt, skill, decision, feedback, session — anything reusable |
| FTS5 search with filters by type, project, tags, status, purpose |
| Retrieve entry by ID with artifact reference |
| Save long AI output as file-backed artifact with metadata |
| Save an AI prompt result as a vault entry |
| Compile agent-ready context pack (7 modes) |
| Get ordered entries in a series |
| Get workflow steps as ordered checklist |
| Create session entry with decisions, pending items, learnings |
| Set entry status to archived |
| List all projects with status |
| Create/update a graph edge between two entries (with cycle detection) |
| List graph edges with filters |
| Traverse entry graph from a starting entry |
| Compare semantic/vector similarity between entries |
| Search entries by tag intersection (all) or union (any) |
| Get structured project context bundle with entries grouped by type |
| Run a workflow with structured step inputs and JSON results |
| Resolve a scenario to a workflow or skill route |
| Return vault statistics including workflow run analytics |
| List workflow runs with optional workflow filter and step progress |
| Get a single workflow run with step details |
| List version history for an entry (descending by version number) |
| Restore an entry to a previous version by version number |
| Save a structured agent-to-agent task handoff / checkpoint |
| Retrieve the latest agent-to-agent task handoff for a project or task_id |
| Store a URL bookmark with an alias keyword for quick retrieval |
| Resolve an alias keyword to its stored URL |
| Search stored URL bookmarks with FTS5 across title, summary, tags, and URL |
| List all stored URL bookmarks, optionally filtered by project |
| Store a command snippet or instruction recipe with placeholders (e.g. |
| Resolve an instruction recipe by alias with optional variable interpolation |
| Search instruction recipes and command snippets with FTS5 |
| List all stored instruction recipes, optionally filtered by project |
| Save typed observation with normalized hash deduplication and topic tags |
| Search observations with FTS5, topic/project filters, and 300-char preview |
| Retrieve full observation details by ID or slug |
| Retrieve chronological sibling observations before and after target ID |
| Compile agent-ready project context pack from active memories |
Engram Drop-in Compatibility Mode (--compat-engram)
To run SkillVault as a 100% drop-in replacement for Engram (exposing mem_save, mem_search, mem_context, mem_get_observation, mem_timeline aliases with zero prompt changes):
skillvault mcp --compat-engramOr set the environment variable SKILLVAULT_COMPAT_ENGRAM=1.
Sovereign Episodic Memory CLI (skillvault mem)
Manage episodic project memories, architectural decisions, and agent context directly from the terminal without requiring an external MCP server:
# Save an observation with automatic normalized hash deduplication
skillvault mem save "Chose SQLite WAL" --content "Enables concurrent readers without blocking writes." --type decision --project myapp --topic architecture/db
# Save from piped stdin or positional text
echo "Learned that Go flag.FlagSet stops on the first positional arg." | skillvault mem save "Go flag parsing caveat" --type discovery --project myapp
# Search observations using FTS5 (sanitized against hyphens and operators)
skillvault mem search "SQLite WAL" --project myapp
# Retrieve full observation details by ID
skillvault mem get obs-1234567890ab
# View chronological context surrounding an observation
skillvault mem timeline obs-1234567890ab --before 2 --after 2
# Compile an agent-ready context pack of recent active memories
skillvault mem context --project myappMCP Setup (Claude Code / OpenCode / Antigravity)
Add to your MCP configuration:
{
"mcpServers": {
"skillvault": {
"command": "/path/to/skillvault",
"args": ["mcp"]
}
}
}Create a symlink for agent-only access:
ln -sf ~/tools/skillvault ~/tools/mcp
# Now agents can call "mcp" directlyEntity Model
Entry Types (15)
Type | Purpose |
| Reusable prompt template |
| Structured skill/pattern |
| Note about a workflow |
| Reference document |
| User preference |
| User feedback/decision |
| Project snapshot |
| Session summary |
| Architectual decision |
| Summary of a stored artifact |
| Session handoff document |
| Actionable task or todo item |
| Scenario → workflow/skill routing rules |
| URL bookmark with alias and tags |
| Command snippet / recipe with variable placeholders |
Purpose Taxonomy
Purpose is orthogonal to entry type. Use it to organize memory by why it exists, not just what shape it has.
Purpose | Use for |
| Active projects, workflows, tasks, deliverables |
| Concepts, references, reusable technical facts |
| Lessons, skill development, retrospectives |
| People, organizations, stakeholder context |
| Current state snapshots, project status, handoffs |
| Logs, metrics, monitoring dashboards, workflow analytics |
Examples:
skillvault add-entry --title "ISO checklist" --type reference --purpose KNOWLEDGE --summary "..."
skillvault search "ISO" --purpose KNOWLEDGEStatus Model
Status | Meaning |
| Not ready for agent use |
| Available for normal retrieval |
| Searchable but excluded from context packs |
| Kept for history, not recommended |
| Preferred version |
Relations (EntryLink — 11 types)
Entries can be explicitly related via a directed graph with cycle detection:
references → links to supporting content
supersedes → newer version replaces older (cycle-prone)
related_to → loosely connected
part_of → compositional relationship (cycle-prone)
derived_from → source → derived relationship
implements → implements a spec/decision
uses → source invokes target (agent uses workflow)
extends → source specializes target
handoff_of → handoff entry refers to session work
generated_from → entry derives from another
depends_on → source depends on target (cycle-prone)Cycle detection: depends_on, part_of, and supersedes validate against
transitive cycles using WITH RECURSIVE CTE before insertion. Max depth: 10.
Traversal: GetEntryGraph supports multi-direction (outgoing|incoming|both)
traversal with configurable depth. CLI: skillvault graph --entry <id>.
Memory index: Shadow entries from pi-memory-md are automatically linked via
external_ref. Wikilinks ([[target]]) in markdown bodies become related_to edges.
Security
SkillVault detects and blocks secrets before they enter the vault:
Pattern | Example |
OpenAI keys |
|
Private keys |
|
GitHub tokens |
|
Slack tokens |
|
On detection: content is rejected or redacted with a warning. No secrets stored.
User Flows
1. Save a PDF analysis from an AI agent
User → "Guarda este análisis en SkillVault como artefacto del proyecto Forense Digital"
Agent → calls save_artifact(title, type=pdf_analysis, content, project)
SkillVault → stores file in objects/2026/06/forense-analisis.md
→ creates artifact metadata + entry in DB
→ indexes summary + tags in FTS52. Agent retrieves planning context
Agent → calls get_context(project="MyApp", mode="planning", max_chars=8000)
SkillVault → compiles: user preferences + project state + decisions + workflows
→ truncates lowest-priority sections to fit 8000 chars
→ returns structured text for agent injection3. Session wrap for project continuity
User → "Cierra sesión, guarda lo que decidimos"
Agent → calls session_wrap(project, summary, decisions[...], pending[...], learnings[...])
SkillVault → creates session entry + optional artifact
→ links to project
→ available for next get_context callComparison
Feature | File system | GitHub Gists | Obsidian | SkillVault |
Agent-facing MCP | ❌ | ❌ | ❌ | ✅ |
FTS5 search | ❌ | ❌ | ❌ | ✅ |
Context compilation | ❌ | ❌ | ❌ | ✅ (Qu@ntum) |
Hybrid DB+disk | ❌ | ❌ | ❌ | ✅ |
Secret detection | ❌ | ❌ | ❌ | ✅ |
Workflow checklists | ❌ | ❌ | ❌ | ✅ |
Workflow pipelines | ❌ | ❌ | ❌ | ✅ (v3) |
Single 10MB binary | ❌ | ❌ | ❌ | ✅ |
Zero cloud dependency | ✅ | ❌ | ❌ | ✅ |
Portable (any OS) | ✅ | ❌ | ❌ | ✅ |
Entry relations | ❌ | ❌ | ❌ | ✅ |
Testing
# Run all 750+ tests
go test ./...
# With coverage
go test -cover ./...
# Integration tests use in-memory SQLite (no filesystem needed)Test pyramid:
Unit tests: domain validation, security patterns, variable detection
Integration tests: SQLite stores, artifact filesystem, MCP tools
Acceptance tests: full end-to-end flows (AC1-AC10)
Dependencies
Go 1.26+ — standard library for CLI, HTTP, JSON, file I/O, crypto, embed
modernc.org/sqlite— pure Go SQLite driver (no CGO)telemetryd— Unix socket daemon for agent observability (included in repo)telemetryctl— CLI for run queries and daemon status (included in repo)
Zero frameworks. No Cobra, no Fiber, no ORM, no Gin, no Echo.
Project Status
Phase | Status |
v1-alpha (SQLite vault) | ✅ Archived |
v2 Qu@ntum (Hybrid + Context) | ✅ Archived |
v3 Qu@ntum (Workflow Pipelines) | ✅ Active |
v3 Service hardening (auth, shutdown, MCP tools) | ✅ Active |
Cloud sync (S3 + GitHub transports) | ✅ Active |
TUI (Bubble Tea, build-tag gated) | ✅ Active |
Vector search (GloVe, pure Go) | ✅ Active |
Entry diff / compare-entries | ✅ Active |
Entry versioning (history, restore, diff API) | ✅ Active |
Entry skill pack export | ✅ Active |
HTTP auth layer | ✅ Active |
Graceful shutdown | ✅ Active |
save_result MCP tool | ✅ Active |
Workflow-builder YAML import | ✅ Active |
Scenario routing ( | ✅ Active |
LifeOS purpose taxonomy | ✅ Active |
Structured MCP workflow runs ( | ✅ Active |
Workflow run analytics ( | ✅ Active |
Agent telemetry daemon ( | ✅ Active |
Agent telemetry CLI ( | ✅ Active |
Agent telemetry plugins + wrapper | ✅ Active |
Quality signal detectors (loop, stall, streak, token) | ✅ Active |
Link bookmarks & quick paste ( | ✅ Active |
Instruction recipes & vars interpolation ( | ✅ Active |
Bidirectional sync & merge ( | ✅ Active |
Author & License
Author: Gabriel Magallon Sanchez - Qu@antum
Organization: QuantumEdu
License: MIT — Copyright (c) 2026 Gabriel Magallon Sanchez - Qu@antum. All rights reserved.
Available Tools
39 toolsadd_instructionC
Store a command snippet or instruction recipe with placeholders (e.g. {{input}})
Inputs: tags, alias, title, command, +2 more.
add_linkC
Store a URL bookmark with an alias keyword for quick retrieval
Inputs: url, tags, alias, title, +2 more.
archive_entryB
Soft-delete an entry (set status to archived)
Inputs: id.
compare_entriesA
Compute line-based LCS unified diff between two entries
Inputs: id1, id2.
compose_seriesC
Return ordered entries in a series
Inputs: series_id.
get_contextC
Compile agent-ready context pack from vault data
Inputs: mode, query, include, project, +2 more.
get_context_bundleB
Return structured JSON bundle: project info, entries grouped by type, and artifact refs
Inputs: project.
get_entryB
Get a vault entry by ID (includes artifact reference if linked)
Inputs: id.
get_entry_graphC
Traverse entry graph from a starting entry, returning connected nodes and edges
Inputs: entry_id, direction, max_depth, ref_types.
get_handoffA
Retrieve the latest agent-to-agent task handoff for a project or task_id
Inputs: project, task_id.
get_instructionB
Resolve an instruction recipe by alias with optional variable interpolation
Inputs: vars, alias.
get_linkC
Resolve an alias keyword to its stored URL
Inputs: alias.
get_runA
Get a single workflow run with its step details
Inputs: run_id.
get_statsB
Return vault statistics including workflow run analytics
Inputs: no named parameters.
list_entry_refsC
List graph edges between entries
Inputs: source_id, target_id, relation_type, include_archived.
list_entry_versionsB
List version history for an entry (descending by version number)
Inputs: entry_id.
list_instructionsA
List all stored instruction recipes, optionally filtered by project
Inputs: project.
list_linksA
List all stored URL bookmarks, optionally filtered by project
Inputs: project.
list_projectsB
List all projects and their statuses
Inputs: no named parameters.
list_workflow_runsC
List workflow runs with optional workflow filter and step progress
Inputs: limit, workflow_id.
render_workflowB
Return workflow steps as a checklist
Inputs: workflow_id.
restore_entry_versionC
Restore an entry to a previous version by version number
Inputs: entry_id, version_number.
route_scenarioC
Resolve a free-text scenario to a matching workflow or skill
Inputs: scenario.
run_workflowC
Execute a workflow with structured step inputs and return per-step results
Inputs: steps, workflow.
save_artifactC
Save a long AI output or file-backed artifact
Inputs: tags, type, title, content, +3 more.
save_entryC
Save a vault entry (prompt, skill, decision, etc.)
Inputs: body, tags, type, title, +4 more.
save_entry_refC
Create or update a link between two entries (graph edge)
Inputs: label, source_id, target_id, relation_type.
save_handoffC
Save a structured agent-to-agent task handoff / checkpoint
Inputs: project, task_id, next_steps, step_summary, +2 more.
save_resultC
Save an AI prompt result as a vault entry
Inputs: name, tags, type, model, +4 more.
search_by_tagsC
Search entries by tags with all/any match against the junction table
Inputs: tags, type, limit, match, +1 more.
search_entriesC
Search entries with FTS5 and filters
Inputs: tags, type, limit, query, +4 more.
search_instructionsC
Search instruction recipes and command snippets with FTS5
Inputs: limit, query, project.
search_linksA
Search stored URL bookmarks with FTS5 across title, summary, tags, and URL
Inputs: limit, query, project.
session_wrapC
Save a compact session summary with decisions, pending items, and learnings
Inputs: pending, project, summary, artifacts, +2 more.
vault_mem_contextC
Build a compact agent-ready memory context pack for a project
Inputs: project.
vault_mem_getC
Get complete content and metadata of an observation memory by ID
Inputs: id, observation_id.
vault_mem_saveA
Save an episodic observation memory with deduplication and topic indexing
Inputs: tags, type, scope, title, +4 more.
vault_mem_searchB
Search observation memories with preview truncation and filters
Inputs: type, limit, query, topic, +2 more.
vault_mem_timelineC
Retrieve chronological context before and after a memory observation
Inputs: id, after, before, project, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 39 tools
Tools are broadly separated by resource (entries, links, instructions, memories, workflows), but there is real overlap: get_context, get_context_bundle, and vault_mem_context all build agent context; save_entry and save_result both save entries; search_entries and search_by_tags overlap. Descriptions help, but an agent can still hesitate between several similarly scoped tools.
Most names are readable snake_case, often verb_noun (save_entry, get_entry, list_links, search_instructions). However, conventions are mixed: vault_mem_save puts the verb last with a namespace prefix, while session_wrap, compose_series, route_scenario, and render_workflow use less predictable verb/noun shapes.
With 39 tools, the server is clearly overgrown for a single MCP surface, spanning entries, graph refs, versions, workflows, runs, memories, links, instructions, handoffs, artifacts, and projects. This likely exceeds what an agent can comfortably scan and select from without grouping or splitting.
Entry lifecycle coverage is strong, including save/get/search/archive, versions, restore, graph refs, tags, compare, and context packs. But secondary resources have notable gaps: links and instructions lack update/delete, memories lack update/delete, handoffs have only save/get, and projects have list but no create/update operations.
Maintenance
Related MCP Connectors
Agent-native notes, tasks, dev-docs, vaults, sync & handoffs. MCP + OpenAPI dual surface.
Persistent memory, hybrid search and a goal graph for AI agents, over stdio or remote HTTP.
Private-by-default, local-first memory/context/task orchestrator for MCP apps and agents.
Self-hosted AI workspace: semantic RAG document search, long-term memory, and AI chat via MCP.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceDeveloper Knowledge OS — local-first bimodal workspace for humans (TUI) and AI agents (MCP). Manage Git repos, vault notes, and assets with 19 MCP tools including unified project context queries.AGPL 3.0

ellmos-homebase-mcpofficial
AlicenseCqualityBmaintenanceEnables local-first LLM orchestration with persistent memory, knowledge management, routing, swarm patterns, API probing, tests, automation planning, and plugin discovery via a stdio MCP server, using SQLite for offline storage.51173 npm2MIT- FlicenseNot gradedqualityBmaintenanceA local-first MCP server and CLI that gives coding agents structured project memory, task contracts, context packs, backlog workflows, and verification evidence, storing data in reviewable Markdown/YAML with a fast SQLite index.4-
- AlicenseNot gradedqualityBmaintenanceExposes a living systems memory vault to autonomous agents, letting them search notes with BM25 ranking, read sections within token budgets, extract API/port/schema contracts, compute dependency blast radius, and run machine-verifiable invariant assertions before mutating code. It also supports surgical frontmatter edits, schema linting, wikilink integrity checks, and worklog recording, over stdio MCP.MIT