Skip to main content
Glama

Project Guardian MCP

MCP server: hela-genome · Version: 3.0.6 · Canonical tools: 31. Current tool reference.

Part of the HeLa MCP Ecosystem — This server is HeLa Genome (hela-genome) — the State & Memory component of the HeLa cellular architecture. See the ecosystem docs for profiles, workflows, and multi-client setup.

A Model Context Protocol (MCP) server for persistent project memory, knowledge-graph operations, SQLite data access, runtime security checks, and guided project-management workflows. The current registry exposes 29 canonical tools, 11 resources, and 27 prompts.

Blotcat — guardian on duty, wiring the knowledge graph from memory.db

Table of Contents

Related MCP server: Engram

Features

Project Guardian Memory System

Blotcat pouring a small project memory bucket into a large central memory vat

  • Knowledge Graph: Maintain project entities, relationships, and observations

  • Entity Management: Projects, tasks, people, resources with rich metadata

  • Relationship Mapping: Dependencies, ownership, blockers, and connections

  • Observation Tracking: Contextual notes and progress updates

  • Progressive Disclosure Retrieval: search_nodes supports tiered retrieval via detailLevel: compact | summary | full. Compact mode returns IDs, types, observation counts, and 1-line previews (~50 tokens) to prevent context window bloat, followed by on-demand deep recall via open_node.

  • Pre-File Context & Timeline: Call get_file_context before reading or refactoring a file to review prior modifications, linked tasks (implements), and open bugs (affects). Call get_timeline for chronological event feeds.

  • Privacy Guardrails: Built-in stripping of <private>...</private> and <!-- private --> tags before persisting observations or indexing into vector embeddings.

  • Zero-Touch Lifecycle Hooks: Shell/Node scripts under hooks/ (session-start.mjs, post-tool-run.mjs) enable automated context injection and change tracking for Claude Code, Cursor, and Git hooks.

  • Semantic Search: Fast, localized RAG matching via SQLite's native FTS5 extension (MATCH and bm25() ranking) across entity names, types, and observations

  • Per-Project Memory: Each project gets its own memory.db. The server resolves the project root in this order: the GUARDIAN_PROJECT_ROOT environment variable, then the Git toplevel of its working directory, then $XDG_DATA_HOME/project-guardian as a shared fallback outside any Git repository

  • Central Memory Mirror: Every memory write also syncs into one central database at ~/memory/memory.db, giving an aggregated, searchable map across all projects and a fallback when a project database is unavailable. Reads through read_graph and search_nodes merge both stores, with project entries taking precedence

  • Daily Central Backups: On the first sync of each day, the central database is snapshotted with SQLite's online backup API and integrity-checked before publication to ~/memory/backup/ddmmyyyy_memory.db; the seven most recent backups are kept. Legacy ~/memory.db migration uses the same snapshot procedure and retains the original for recovery.

  • On-Demand Pre-Commit Setup: Nothing is installed at startup. Call setup_pre_commit when you want a generated .pre-commit-config.yaml and Git hooks in the active project

  • Optional Obsidian Vault Mirror: When HELA_GENOME_OBSIDIAN_VAULT is set, entity/relation/observation writes are mirrored as Markdown notes with YAML frontmatter into an Obsidian vault, so humans can browse and edit project memory in Obsidian (graph view, backlinks). SQLite remains the base engine; the web UI dashboard has been removed in favor of Obsidian.

Harness Session Bridge

  • Read-only ingestion from 8 agent harnesses: opencode, kilocode, zed, delta, antigravity, cursor, vscode, codex. Sources are opened read-only (SQLite OPEN_READONLY, capped JSONL scans) — the bridge never writes to harness stores and never touches harness processes.

  • What's harvested: session titles, summaries/previews, timestamps, project directories, models, message/step counts. Full transcripts and binaries are never ingested.

  • Secret redaction first: API keys, tokens, bearer credentials, private keys, and password assignments are replaced with [REDACTED] before anything reaches the graph.

  • Incremental + idempotent: per-session watermarks in a harness_sync table — re-runs only pick up new or updated sessions; dryRun previews without writing.

  • Absent stores are honest: harnesses with no local store on the machine (e.g. Cursor/VS Code when not installed) report found: false instead of failing. Store paths are overridable via HARNESS_<ID>_PATH.

Streamlined Database Operations

Blotcat efficiently sorting raw data blocks on a conveyor belt into the structured memory.db SQLite wall

  • Two Stores, One Interface: Every project uses its own memory.db; all seven database tools can also address the central aggregate with database: "central"

  • Core CRUD: Essential database operations (query, insert, update, delete)

  • SQL Execution: Direct SQL query execution

  • Data Transfer: Import/export CSV and JSON files

  • 31 Canonical Tools + Legacy Aliases: Seven database tools, nine memory/guidance tools (initialize_memory, memory_edit, read_graph, read_graph_stream, search_nodes, open_node, get_file_context, get_timeline, get_project_guidance), three fleet mission tools (create_mission, list_missions, update_mission), two search-feedback tools (search_feedback, search_priors), and ten runtime tools (including obsidian_vault). The view-only Web UI dashboard has been removed in favor of the optional Obsidian vault mirror. Legacy names such as create_entity remain as deprecated aliases that normalize to canonical actions.

Runtime Companion Integration

Blotcat acting as a conductor for miniature sub-Blotcats acting as security, memory, and tracker companions

The repository includes six guardian-* AgentSkills and exposes their operational capabilities through typed MCP tools:

Companion

Runtime role

MCP surface

guardian-memory

Persistent entities, relations, and observations

memory_edit, graph/search/context tools

guardian-session

Active-task, bug, blocker, and recent-change summaries

get_session_context

guardian-tracker

Bounded Git diff and untracked-file analysis

analyze_git_changes

guardian-wall

Untrusted-text normalization and prompt-injection detection

inspect_untrusted_text

guardian-security

Secret scanning and Trivy image scanning

security_scan (secrets / container)

guardian-cache

Optional namespaced Redis storage

cache (get / set / delete / scan)

AgentSkills provide host-side workflows and instructions. The MCP runtime implements the corresponding operations directly in TypeScript, except container scanning, which invokes Trivy as a bounded external process. No generic script or shell execution tool is exposed.

AI Guidance System

Blotcat as an academic master pointing at a glowing scroll of strict rules and project prompts

  • 11 Resources: Templates, best practices, project status, and companion capability health

  • 27 Prompts: Comprehensive pre-built workflows for all aspects of project management

  • Expert Guidance: Step-by-step instructions for complex operations

  • Contextual Help: Adaptive prompts based on user needs

  • Knowledge Base: Comprehensive project management wisdom

Advanced Features

  • Schema Validation: Comprehensive input validation with Zod schemas

  • Error Handling: Detailed error messages and graceful failure handling

  • Connection Management: Bounded 20-connection LRU cache with WAL + synchronous=NORMAL + cache_size=-64000 + journal_size_limit=67108864 + temp_store=MEMORY + busy_timeout=5000, maintenance and shutdown cleanup. The removed web UI's REST routes are not a supported interface.

  • File Integration: CSV and SQL imports are streamed; CSV writes use bounded string assembly

  • Result Limits & Pagination: Raw SELECT results are bounded; use the MCP read_graph, read_graph_stream, and search_nodes schemas for supported pagination and search limits, rather than removed UI HTTP routes.

Enterprise Features

  • TypeScript: Fully typed with comprehensive error handling

  • Input Validation: Zod schema validation for all parameters

  • Error Recovery: Graceful error handling with detailed error messages

  • Resource Management: Automatic cleanup of connections and resources

  • Testing: The current isolated regression run passes 19 suites / 264 tests, including vault conflicts, replicated timestamp/observation/relation-order convergence, typed request/import validation, SQL binding validation, WAL-safe backups, mission contention, and vector retrieval. Historical UI tests do not imply an available UI service. npm run lint checks every source TypeScript file and fails on any warning; the current run has zero errors and zero warnings.

Requirements

  • Node.js: Node 22 LTS is the validated ecosystem environment; see package.json for the package's declared standalone engine range.

  • npm: Latest stable version

  • SQLite3: Automatically installed as dependency

  • Redis: Optional; required only for cache actions through REDIS_URL

  • Trivy: Optional; required only for security_scan with action: "container"

Installation

  1. Clone the repository:

git clone https://github.com/Hela-Ecosystem/project-mcp.git
cd project-mcp
  1. Install dependencies:

npm install
  1. Build the project: Choose between development or production build:

For development (includes source maps and full TypeScript compilation):

npm run build

For production (creates an optimized, minified bundle):

npm run build:prod
  1. Run the test suite:

npm test
  1. Start the server:

npm start

Updating After Changes

When you pull new updates or modify the code, you must rebuild the server and restart your MCP client (Cursor, Claude Desktop, etc.) for the changes to take effect:

  1. Pull the latest code: git pull

  2. Install new dependencies (if any): npm install

  3. Rebuild the bundle: npm run build:prod

  4. Important: Restart your IDE or the MCP connection so the client can fetch the newly updated tools and prompts.

Available Tools

Blotcat opening a large toolbox with three labeled drawers, holding a wrench

This MCP server currently advertises 29 canonical tools. The generated tool reference lists their names and consolidated actions.

Tool reference note: sections below preserve legacy entry points for compatibility. The canonical surface is 29 tools (added obsidian_vault and 3 fleet mission tools create_mission, list_missions, update_mission; removed ui_server in favor of the Obsidian vault mirror). Legacy names such as create_entity and cache_get normalize to canonical actions and emit a deprecation warning. Section counts include those aliases.

Database Operations (7 tools)

All database tools accept an optional database selector: project (default) targets the active project's memory.db, central targets the central aggregate at ~/memory/memory.db.

execute_sql - Execute SQL Query

Execute raw SQL queries on the selected memory database.

Parameters:

  • query (required): SQL query string

  • parameters (optional): Query parameters array

  • database (optional): "project" or "central", default "project"

query_data - Query Table Data

Query memory tables with filtering and pagination.

Parameters:

  • table (required): Table name

  • conditions (optional): WHERE conditions object

  • limit (optional): Maximum rows to return

  • offset (optional): Number of rows to skip

  • orderBy (optional): Column to sort by

  • orderDirection (optional): Sort direction ("ASC" or "DESC")

  • database (optional): "project" or "central", default "project"

insert_data - Insert Records

Insert records into a memory table.

Parameters:

  • table (required): Table name

  • records (required): Array of record objects to insert

  • database (optional): "project" or "central", default "project"

update_data - Update Records

Update records in a memory table.

Parameters:

  • table (required): Table name

  • conditions (required): WHERE conditions for records to update

  • updates (required): Fields to update

  • database (optional): "project" or "central", default "project"

delete_data - Delete Records

Delete records from a memory table.

Parameters:

  • table (required): Table name

  • conditions (required): WHERE conditions for records to delete

  • database (optional): "project" or "central", default "project"

import_data - Import Data

Import data from CSV or JSON file into a memory table.

Parameters:

  • table (required): Target table name

  • filePath (required): Workspace-relative path to the source file (absolute paths and .. traversal are rejected unless HELA_GENOME_ALLOW_EXTERNAL_PATHS=true)

  • format (optional): File format ("csv" or "json")

  • options (optional): Import options (delimiter, hasHeader)

  • database (optional): "project" or "central", default "project"

export_data - Export Data

Export memory table data to CSV or JSON file.

Parameters:

  • table (required): Source table name

  • filePath (required): Workspace-relative output path (absolute paths and .. traversal are rejected unless HELA_GENOME_ALLOW_EXTERNAL_PATHS=true)

  • format (optional): Output format ("csv" or "json")

  • conditions (optional): WHERE conditions to filter export

  • options (optional): Export options (delimiter, includeHeader)

  • database (optional): "project" or "central", default "project"

Memory and Guidance Tools

Use memory_edit for entity/relation/observation mutations and harness_sessions with action: "list" or "sync" for harness ingestion. Older entry points documented below are compatibility aliases, not additional advertised tools.

initialize_memory - Initialize Memory System

Set up the project memory database schema and tables.

Parameters: None

create_entity - Create Project Entities

Create entities in the project knowledge graph (supports single or batch).

Parameters:

  • entities (required): Array of entity objects

    • name: Entity name

    • entityType: Type (project, task, person, resource)

    • observations: Array of notes about the entity

create_relation - Create Entity Relationships

Create relationships between project entities (supports single or batch).

Parameters:

  • relations (required): Array of relation objects

    • from: Source entity name

    • to: Target entity name

    • relationType: Relationship type (depends_on, blocks, owns, etc.)

add_observation - Add Entity Observations

Add observations/notes to project entities (supports single or batch).

Parameters:

  • observations (required): Array of observation objects

    • entityName: Target entity name

    • contents: Array of observation strings to add

delete_entity - Delete Project Entities

Remove entities and their relations from project memory (supports single or batch).

Parameters:

  • entityNames (required): Array of entity names to delete

delete_observation - Remove Entity Observations

Remove specific observations from entities (supports single or batch).

Parameters:

  • deletions (required): Array of deletion objects

    • entityName: Target entity name

    • observations: Array of observation strings to remove

delete_relation - Delete Entity Relationships

Remove relationships between project entities (supports single or batch).

Parameters:

  • relations (required): Array of relation objects to delete

    • from: Source entity name

    • to: Target entity name

    • relationType: Relationship type to delete

read_graph - Read Project Knowledge Graph

Retrieve the full knowledge graph, merging the active project database with the central aggregate. Project entries win over central entries with the same name. Supports pagination.

Parameters:

  • database (optional): "project" (default, merged), "central" (only central)

  • limit (optional, 1-10000, default 5000): Max entities/relations to return, ORDER BY updated_at DESC

  • offset (optional, 0+): Rows to skip

search_nodes - Search Project Knowledge (Hybrid RAG & Progressive Disclosure)

Search for entities and relations matching a query across names, types, and content, using SQLite FTS5 (BM25) and local vector embeddings with Reciprocal Rank Fusion (RRF $k=60$). Supports progressive disclosure to prevent token bloat.

Parameters:

  • query (required): Search term

  • limit (optional, 1-100, default 20): Max ranked entities to return

  • mode (optional): "keyword" (FTS5 BM25, default), "vector" (sqlite-vec 384d), or "hybrid" (RRF fusion)

  • detailLevel (optional): Progressive disclosure level:

    • "compact": Returns lightweight IDs, entity types, observation counts, and 1-line previews (~50 tokens each).

    • "summary": Returns last 2 truncated observations.

    • "full": Full entity graph with all observations (default).

open_node - Get Entity Details

Retrieve complete information and all observations about specific project entities (Tier 3 deep recall).

Parameters:

  • names (required): Array of entity names to retrieve

get_file_context - Fast File History & Context Lookup

Pre-flight check before editing or reading a file. Retrieves prior modification observations, linked features/tasks (implements), and linked blockers/bugs (affects / blocks).

Parameters:

  • filePath (required): Workspace-relative or absolute file path (e.g. "src/auth/jwt.ts")

get_timeline - Chronological Activity Feed

Returns a chronological stream of project observations and events sorted newest first.

Parameters:

  • since (optional): ISO date string or timestamp filter

  • entityType (optional): Filter by entity type (task, bug, file, decision, project)

  • limit (optional, 1-100, default 20): Max events to return

list_harness_stores - List Harness Session Stores

List local agent-harness session stores (opencode, kilocode, zed, delta, antigravity, cursor, vscode, codex): paths, presence, and format notes. Read-only.

Parameters: None

sync_harness_sessions - Sync Harness Sessions Into Memory

Read-only sync of harness sessions (titles, summaries, timestamps; secrets redacted) into the knowledge graph as session entities named harness:<harness>:<sessionId>. Incremental via watermarks.

Parameters:

  • harnesses (optional): Subset to sync (default: all found)

  • project (optional): Substring filter on session project/directory

  • since (optional): Only sessions updated after this ms epoch

  • limit (optional, 1-500, default 50): Max sessions per harness

  • dryRun (optional): Report what would sync without writing

  • includeArchived (optional): Include archived sessions (default false)

get_project_guidance - Access AI Guidance

Invoke a project guidance framework to receive specialized instructions and checklists for specific workflows. This allows the AI to autonomously fetch and follow established project management protocols.

Parameters:

  • guidance_name (required): Name of the guidance (e.g., project-setup, sprint-planning)

  • arguments (optional): Arguments required by the specific guidance framework

Fleet Mission Ledger Tools (3 tools)

Coordinate agents through a SQLite-backed mission board. Claims are exclusive across connections and processes using the same SQLite database. Agents on different machines must call the same server/ledger endpoint to coordinate; separate local databases, central graph synchronization, and vault synchronization do not provide a distributed lock or replicate missions.

create_mission

Enqueue a mission into the fleet mission ledger. Supports assigning to specific agent targets (frea, mema, orchestrator, human, or any), priorities (low, normal, high, urgent), and arbitrary JSON payloads.

Parameters:

  • title (required): Summary or title of the mission

  • description (optional): Detailed context or instructions

  • originAgent (optional, default: agent): Originating agent or user

  • targetAgent (optional, default: any): Target agent assigned to execute

  • priority (optional, default: normal): low | normal | high | urgent

  • payload (optional): Input parameters or data

  • database (optional, default: project): project | central

list_missions

Query missions from the ledger filtered by status, agent target, origin, or claiming agent.

Parameters:

  • id (optional): Specific mission ID to fetch

  • status (optional): queued | claimed | in_progress | completed | failed | cancelled

  • targetAgent (optional): Filter by target agent

  • originAgent (optional): Filter by origin agent

  • claimedBy (optional): Filter by agent currently executing

  • limit (optional, default: 50, max: 200): Pagination limit

  • offset (optional, default: 0): Pagination offset

  • database (optional, default: project): project | central

update_mission

Claim a queued, unowned mission or report its lifecycle progress (e.g. marking completed with output results, or failed with an error message). Automatically records timestamps (claimed_at, completed_at, updated_at). A claim uses one conditional SQL update; competing claimants receive an error rather than replacing the winner.

Parameters:

  • id (required): Mission ID to update

  • status (optional): New status

  • claimedBy (optional): Non-empty agent name, up to 100 characters. Supplying it attempts an exclusive claim, optionally with status: claimed or status: in_progress. Do not repeat it on subsequent lifecycle updates; claimant reassignment and claimant-free status: claimed are rejected.

  • result (optional): JSON result or return value upon completion

  • error (optional): Error message if mission failed

  • database (optional, default: project): project | central

Runtime Companion Tools

The canonical wrappers are security_scan and cache. The scan_* and cache_* entry points below are legacy aliases; use the action-based wrappers in new integrations.

sync_central_memory

Merge the active project knowledge graph into the central memory database (~/memory/memory.db by default, override with GUARDIAN_CENTRAL_DB). Observation unions preserve independent facts and their existing order; entity type follows the newer timestamp, and relations are unique by endpoints/type. Central imports use the same additive merge instead of overwriting local observations. Every memory write also syncs automatically; call this tool to force a sync on demand. The first sync of each day also snapshots the central database and prunes old backups beyond the newest seven. Central memory remains an aggregate, not a deletion-propagating replica or a mission ledger synchronizer.

set_project_root

Switch the active project memory database to the given absolute project path. Use this at session start when the server was launched outside the project directory, so memory is written to the project instead of the shared fallback database.

  • path (required): Absolute path to the project root. Inside a Git repository, the toplevel is used.

setup_pre_commit

Create a .pre-commit-config.yaml in the active project root and install the Git hooks, on demand. Requires pre-commit to be installed. The generated .gitignore entries are intentionally broad: alongside memory.db, the block ignores common local tool directories such as .claude/, .vscode/, .idea/, .gemini/, and .cursor/, plus .env files. Entries already present in .gitignore are never duplicated. The server never does any of this automatically at startup.

get_session_context

Summarize active tasks, open bugs, recent changes, blockers, and the next suggested action directly from the knowledge graph.

  • limit (optional, 1-50, default 10): Maximum entries per result group.

analyze_git_changes

Return exact machine-readable changed paths from Git, including renames and optionally untracked files.

  • commit (optional): Analyze one commit against its parent.

  • since (optional, default 1): Analyze changes since N commits ago or a Git date.

  • includeUntracked (optional, default true): Include untracked files for worktree analysis.

  • maxFiles (optional, 1-500, default 100): Bound returned paths.

  • commit and a custom since value are mutually exclusive.

inspect_untrusted_text

Normalize up to 256 KiB of untrusted text and detect hidden formatting, instruction overrides, role mimicry, hidden HTML/CSS, remote exfiltration markup, and encoded instruction-like content.

  • text (required): External or otherwise untrusted content.

  • Detection is heuristic. Returned normalized text remains untrusted data.

scan_project_secrets

Scan a workspace-relative file or directory for likely hardcoded credentials. Results contain only type, relative file path, and line number; matched values are never returned.

  • path (optional, default .): Workspace-relative scan target.

  • exclude (optional): Additional directory names to skip.

  • maxFindings (optional, 1-500, default 100): Bound findings.

  • Absolute paths, traversal, missing paths, and symlink escapes are rejected.

scan_container_image

Run a timeout-limited Trivy scan and return bounded HIGH/CRITICAL vulnerability summaries.

  • image (required): Container image reference.

  • maxFindings (optional, 1-500, default 100): Bound findings.

  • Requires Trivy. Image values beginning with -, containing whitespace, or containing control characters are rejected.

Redis cache tools

  • cache_get: Read one mema:<category>:<name> key.

  • cache_set: Store a value up to 512 KiB with optional ttlSeconds from 1 to 604800.

  • cache_delete: Delete one namespaced key.

  • cache_scan: Cursor-scan a mema:* pattern with a bounded count.

Project scan paths are restricted to the current Git workspace. Redis tools connect lazily and return an unavailable error when REDIS_URL is unset. Container scanning remains unavailable until Trivy is installed. Read project-guardian://companions/catalog for current capability health.

Obsidian Vault Mirror (optional, remote-capable, agent-controllable)

Set HELA_GENOME_OBSIDIAN_VAULT=/path/to/vault to enable the optional Markdown mirror. SQLite remains the base engine; no Obsidian service is required. Startup imports managed entity and relation notes without clearing SQLite. Only bullets in ## Observations are facts; generated ## Relations backlinks are never imported as observations. Repeated imports are idempotent and retain junk/test-memory guards.

MCP tool: obsidian_vault supports status, sync, rebuild, backfill, sync_central, sync_from_central, and dedup. Rebuild imports notes, not a live database-file replacement. Local deduplication preserves edges/provenance and remembers aliases so central imports do not recreate merged names.

The watcher polls every two seconds and serializes imports. Human observation changes replace the rendered baseline only if SQLite still matches it; divergent agent/human edits merge their observations. Previous note revisions are retained under .hela-conflicts/. Hidden directories are excluded from imports. Deletions are inferred only for notes previously observed by that running watcher, using frontmatter identity; a missing or unreadable vault fails closed. Absence on startup is not a deletion instruction. Central memory is additive: a later explicit central import may restore a project-deleted entity.

Replicated notes retain the earliest creation date and latest update date instead of rewriting remote timestamps to local ones. A replicated observation superset can retain its incoming order without dropping database-only facts. Relation links use a canonical order independent of database insertion order. These rules prevent timestamp/order-only sync loops; genuinely divergent notes still require a deliberate resolution.

Conflict-aware synchronization between machines

Build with npm run build, then configure both legacy hook names to the same two-way transport. The pull hook runs asynchronously at startup and before explicit sync; the write hook is debounced and serialized. Hook failures reach explicit callers; background failures are logged without printing command contents. Legacy rsync --delete hooks are rejected.

export HELA_GENOME_OBSIDIAN_VAULT=/path/to/local/vault
export HELA_GENOME_OBSIDIAN_PULL_CMD="node /path/to/project-mcp/dist/vault-sync.js --vault /path/to/local/vault --peer /path/to/remote/vault --remote user@host"
export HELA_GENOME_OBSIDIAN_SYNC_CMD="$HELA_GENOME_OBSIDIAN_PULL_CMD"

Both directories must already exist. The peer requires SSH access and Node.js, but not an installed Genome server or helper. Omit --remote to test two local vault directories.

  • First sync unions missing files; it never treats initial absence as deletion.

  • A shared hash baseline permits one-sided edits and deletions only when the other side is unchanged. Divergent edits and edit/delete conflicts leave both canonical states untouched and archive both available revisions plus deletion intent in .hela-sync/conflicts/ on both machines.

  • Overwritten or deleted revisions are retained in .hela-sync/recovery/. Archives contain .bin content plus JSON path/hash records. Resolve conflicts by copying the chosen/merged content to both canonical paths, or deleting both copies, then rerun sync.

  • Programmatic repairs can opt into syncVaults(..., { resolveConflict }). Return a Buffer containing reviewed merged content, or undefined to leave the conflict unresolved. Both originals are archived before publication. beforeApply(result) runs under both locks and can reject the complete copy/delete/conflict plan before canonical files change. The CLI does not enable a resolution policy automatically.

  • Exit codes: 0 synchronized, 2 unresolved conflicts, 1 failed/unavailable. Missing peers, corrupt metadata, and failed SSH commands are not empty snapshots.

  • Locks serialize cooperating synchronizers; expected-hash checks reject stale snapshots. Editors do not honor these locks, so do not treat this as a distributed filesystem transaction. Interrupted or mismatched baselines fall back to conservative reconciliation. Recovery archives preserve replaced data.

  • Non-hidden regular files, including attachments, are managed; .obsidian/, .hela-sync/, and other hidden paths are excluded. Symlinks, hardlinks, overlapping local roots, and oversized snapshots are rejected. Default transport limit is 32 MiB per operation with a conservative snapshot budget; the exported API accepts up to 64 MiB.

  • Keep each vault's .hela-sync/id and state local; do not clone these metadata files between peers. A stale .hela-sync/lock fails closed. Remove only that specific lock after confirming its owning sync process is no longer running; do not erase the baseline to force a deletion.

Run npm test -- --runInBand for isolated coverage. node scripts/test-vault-sync-ssh.mjs user@host checks create/edit/delete conflicts and recovery using disposable directories on both hosts; it never uses the configured production vault.

SQLite backup and isolated restore

SQLiteManager.backupDatabase(database, backupPath) uses SQLite's online backup API, includes committed WAL data, verifies the SQLite header and PRAGMA integrity_check, and publishes a closed snapshot atomically without overwriting an existing destination. Daily central backups and legacy migration use this implementation.

SQLiteManager.restoreDatabase(backupPath, newDatabaseNameOrAbsolutePath) restores only to a new destination after validation. Existing database files, active connections, aliases to existing targets, and SQLite sidecars are refused. A successful isolated restore is evidence that the backup is readable; it is not authorization to overwrite production storage.

For production recovery: stop every writer, take a fresh recovery backup, restore under a new name, verify integrity and expected records, then explicitly switch the service configuration to that restored database. Retain the old database and its WAL/SHM as a coordinated offline recovery set. Never copy over an active SQLite database or delete its sidecars to force a restore.

AI Guidance System

Project Guardian MCP includes comprehensive resources and prompts to help AI models effectively use the toolset for project management.

Available Resources

Project Guardian provides 11 key resources that AI models can read to understand project management concepts, access capability health, and get comprehensive project insights:

project-guardian://templates/entity-types

Standard entity types for project management with examples and usage guidelines.

project-guardian://templates/relationship-types

Common relationship types between project entities with practical examples.

project-guardian://templates/project-workflows

Standard workflows for using Project Guardian tools in different scenarios.

project-guardian://templates/best-practices

Comprehensive best practices guide for effective project knowledge management.

project-guardian://status/current-graph

Current state of the project knowledge graph with summary statistics.

project-guardian://cache/recent-activities

Recently performed project management activities and updates for tracking progress.

project-guardian://cache/workflow-templates

Frequently used workflow templates with examples and implementation guidance.

project-guardian://metrics/project-stats

Statistical overview of project entities, relationships, and activities with health metrics.

project-guardian://cache/team-members

Cached information about project team members and their roles within the organization.

project-guardian://status/recent-changes

Recent additions, updates, and modifications to the knowledge graph for audit and monitoring.

project-guardian://companions/catalog

Lists all six companions, their MCP tools, external prerequisites, and current availability.

Available Prompts

Project Guardian offers 27 prompts covering project setup, planning, quality, operations, and incident workflows:

Core Project Management

project-setup - Project Initialization

Arguments:

  • project_name (required): Name of the project

  • team_members (optional): Comma-separated list of team members

Provides step-by-step guidance for setting up a new project structure with appropriate entities and relationships.

sprint-planning - Sprint Planning

Arguments:

  • sprint_name (required): Name/number of the sprint

  • duration_days (optional): Sprint duration in days

Guides through comprehensive sprint planning including task breakdown, dependencies, and capacity planning.

progress-update - Progress Tracking

Arguments:

  • task_name (required): Name of the task to update

  • progress_notes (required): Progress update description

Structured process for updating task progress and managing dependencies.

retrospective - Project Retrospective

Arguments:

  • time_period (required): Time period being reviewed (e.g., "last sprint", "Q1")

Comprehensive retrospective process including data analysis, pattern identification, and improvement action creation.

Quality & Process Management

code-review - Code Review Process

Arguments:

  • pull_request_title (required): Title of the pull request being reviewed

  • reviewer_name (optional): Name of the reviewer

Structured code review process with technical checklists, issue documentation, and approval workflows.

bug-tracking - Bug Management

Arguments:

  • bug_description (required): Description of the bug or issue

  • severity_level (optional): Critical, High, Medium, or Low severity

Complete bug tracking workflow from discovery to resolution with impact analysis and stakeholder communication.

technical-debt-assessment - Technical Debt Analysis

Arguments:

  • component_name (required): Name of the component or codebase being assessed

  • assessment_scope (optional): Scope of assessment (file, module, system)

Comprehensive technical debt identification, prioritization, and remediation planning.

Release & Deployment Management

release-planning - Release Planning

Arguments:

  • release_version (required): Version number for the release (e.g., "v2.1.0")

  • release_date (optional): Target release date

Complete release planning process including quality gates, risk assessment, and deployment coordination.

Risk & Change Management

risk-assessment - Risk Management

Arguments:

  • risk_description (required): Description of the risk

  • impact_level (optional): High, Medium, or Low impact

Complete workflow for documenting risks, identifying impacts, and developing mitigation strategies.

change-management - Change Control

Arguments:

  • change_description (required): Description of the proposed change

  • impact_assessment (optional): High, Medium, or Low impact assessment

Structured change management process with impact analysis, approval workflows, and implementation tracking.

Team & Resource Management

team-productivity - Productivity Analysis

Arguments:

  • timeframe (required): Time period to analyze (week, month, quarter)

  • focus_area (optional): Area to focus on (velocity, quality, collaboration)

Team productivity assessment with performance metrics, root cause analysis, and improvement planning.

resource-allocation - Resource Planning

Arguments:

  • resource_type (required): Type of resource (human, infrastructure, budget)

  • planning_horizon (optional): Planning timeframe (sprint, quarter, year)

Resource allocation optimization with capacity planning, gap analysis, and utilization tracking.

Documentation & Communication

stakeholder-communication - Communication Management

Arguments:

  • communication_type (required): Type of communication (status_update, issue_alert, milestone_reached)

  • audience (optional): Target audience (team, management, client, all)

Stakeholder communication planning and execution with audience-specific strategies and effectiveness tracking.

documentation-management - Documentation Updates

Arguments:

  • documentation_type (required): Type of documentation (api, user_guide, technical_spec)

  • update_reason (optional): Reason for documentation update

Documentation maintenance process with content planning, review workflows, and publishing coordination.

Requirements & Planning Management

requirements-gathering - Requirements Gathering

Arguments:

  • requirement_type (required): Type of requirements (functional, non-functional, business, technical)

  • stakeholders (optional): Comma-separated list of key stakeholders

Guides through comprehensive requirements gathering process with stakeholder management and requirement categorization.

user-story-management - User Story Management

Arguments:

  • feature_name (required): Name of the feature or epic

  • user_role (optional): Primary user role (e.g., "customer", "admin", "developer")

Structured process for creating, managing, and prioritizing user stories with acceptance criteria and dependencies.

Quality & Technical Management

testing-strategy - Testing Strategy Development

Arguments:

  • application_type (required): Type of application (web, mobile, api, desktop)

  • criticality_level (optional): Business criticality (critical, high, medium, low)

Comprehensive testing strategy development including automated testing, quality gates, and risk-based testing.

security-assessment - Security Assessment

Arguments:

  • assessment_scope (required): Scope of security assessment (application, infrastructure, data)

  • compliance_requirements (optional): Compliance standards (GDPR, HIPAA, SOC2, etc.)

Security assessment framework with vulnerability management, compliance verification, and security controls implementation.

performance-optimization - Performance Optimization

Arguments:

  • performance_metric (required): Primary metric to optimize (response_time, throughput, resource_usage)

  • optimization_goal (optional): Specific performance target or improvement percentage

Performance monitoring setup, bottleneck identification, and optimization implementation with continuous monitoring.

ci-cd-setup - CI/CD Pipeline Setup

Arguments:

  • pipeline_type (required): Type of pipeline (build, test, deploy, full_ci_cd)

  • target_platform (optional): Deployment target (aws, azure, gcp, kubernetes, heroku)

Complete CI/CD pipeline setup including quality gates, rollback procedures, and security integration.

architecture-review - Architecture Review

Arguments:

  • architecture_type (required): Type of architecture (microservices, monolithic, serverless, hybrid)

  • review_focus (optional): Primary focus area (scalability, security, maintainability, performance)

Architectural assessment framework with design pattern analysis, technology stack evaluation, and improvement recommendations.

Knowledge & Team Management

knowledge-transfer - Knowledge Transfer

Arguments:

  • knowledge_domain (required): Domain of knowledge (technical, process, business)

  • transfer_recipients (optional): Who needs to receive the knowledge (team, individual, department)

Knowledge transfer planning and execution with session management, documentation, and effectiveness validation.

vendor-management - Vendor Management

Arguments:

  • vendor_type (required): Type of vendor service (cloud, development, consulting, infrastructure)

  • contract_value (optional): Contract value range (small, medium, large, enterprise)

Vendor relationship management including contract tracking, performance monitoring, and cost optimization.

Incident & Crisis Management

incident-response - Incident Response

Arguments:

  • incident_severity (required): Severity level (critical, high, medium, low)

  • incident_type (optional): Type of incident (security, performance, functionality, availability)

Incident response framework with containment, recovery, root cause analysis, and post-incident review.

Financial & Resource Management

cost-management - Cost Management

Arguments:

  • cost_category (required): Primary cost category (infrastructure, personnel, tools, licenses)

  • budget_constraint (optional): Budget constraint level (strict, flexible, unlimited)

Cost monitoring, optimization strategies, and budget management with forecasting and reporting.

Customer & Innovation Management

customer-feedback - Customer Feedback Management

Arguments:

  • feedback_channel (required): Primary feedback channel (survey, support, reviews, analytics)

  • feedback_focus (optional): Focus area (usability, features, performance, support)

Customer feedback collection, analysis, and action planning with continuous improvement cycles.

innovation-planning - Innovation Planning

Arguments:

  • innovation_type (required): Type of innovation (product, process, technology, business_model)

  • risk_tolerance (optional): Risk tolerance level (conservative, moderate, aggressive)

Innovation management framework with idea generation, experimentation, and success measurement.

How AI Models Use Guidance

  1. Discovery: List available resources and prompts to understand capabilities

  2. Learning: Read relevant resources to understand project management concepts

  3. Planning: Use appropriate prompts for complex workflows

  4. Execution: Follow structured guidance to use tools effectively

  5. Verification: Check results and iterate as needed This guidance system ensures AI models can provide expert-level project management assistance using the Project Guardian toolset.

Behavioral Protocol (System Rules)

Every prompts/get response from this MCP server includes a shared Behavioral Protocol as a system message (implemented in src/prompts/behavioral-protocol.ts). This protocol enforces:

  • Minimal, production-ready, self-documenting code with a security-first approach.

  • No buzzwords, unnecessary emoji, or filler; direct, technically accurate answers.

  • Adaptive response depth based on the user's request (quick answers vs. complex breakdowns).

  • Consistent use of validated best practices for systems, programming, UI/UX, and design.

Clients integrating this MCP server should treat the first system message as the governing rules for any downstream model that uses these prompts.

Usage Examples

Blotcat routing prompts and tools into memory.db

Project Guardian Setup

// Initialize the project memory system
const initResult = await mcpClient.callTool('initialize_memory', {});

// Create your first project entities
const entityResult = await mcpClient.callTool('create_entity', {
  entities: [
    {
      name: 'web_platform',
      entityType: 'project',
      observations: ['Main web application platform', 'React + Node.js stack', 'Q2 2024 delivery']
    },
    {
      name: 'user_authentication',
      entityType: 'feature',
      observations: ['OAuth2 implementation', 'Google/GitHub providers', 'JWT tokens']
    }
  ]
});

// Establish project relationships
const relationResult = await mcpClient.callTool('create_relation', {
  relations: [
    {
      from: 'user_authentication',
      to: 'web_platform',
      relationType: 'part_of'
    }
  ]
});

Project Management Workflow

// Add progress observations
await mcpClient.callTool('add_observation', {
  observations: [
    {
      entityName: 'user_authentication',
      contents: [
        'Completed OAuth2 setup for Google provider',
        'JWT implementation finished',
        'Unit tests passing at 95% coverage'
      ]
    }
  ]
});

// Search project knowledge
const searchResult = await mcpClient.callTool('search_nodes', {
  query: 'authentication'
});

// Read entire project knowledge graph
const graphResult = await mcpClient.callTool('read_graph', {});

// Get detailed entity information
const entityDetails = await mcpClient.callTool('open_node', {
  names: ['user_authentication', 'web_platform']
});

Database Operations

// Execute custom SQL queries
const sqlResult = await mcpClient.callTool('execute_sql', {
  query: 'SELECT * FROM entities WHERE entity_type = ?',
  parameters: ['project']
});

// Query project data
const queryResult = await mcpClient.callTool('query_data', {
  table: 'entities',
  conditions: { entity_type: 'task' },
  limit: 10
});

// Import/export data
const importResult = await mcpClient.callTool('import_data', {
  table: 'project_data',
  filePath: './project_backup.csv',
  format: 'csv'
});

Configuration

Blotcat plugging a giant power cord into a wall socket

Environment Variables

The server reads these variables at startup:

Variable

Default

Purpose

GUARDIAN_PROJECT_ROOT

unset

Absolute path to the project root. When set, memory.db is stored here instead of relying on Git detection

GUARDIAN_CENTRAL_DB

~/memory/memory.db

Absolute path to the central memory database that every project syncs into. Backups are written to a backup/ directory next to it

GUARDIAN_AUTO_MERGE

unset

Set to 1 to enable scattered-database consolidation at startup. This merges nested memory.db files into the project-root database and deletes them, so leave it unset when sub-projects keep separate memories

REDIS_URL

unset

Enables the Redis-backed cache_* tools

XDG_DATA_HOME

platform default

Base directory for the shared fallback database outside a Git repository

HELA_GENOME_ALLOW_SQL_WRITE

unset = off

Set to true to allow execute_sql writes. Read-only SELECT/WITH/VALUES/EXPLAIN and whitelisted PRAGMA reads always allowed.

HELA_GENOME_ALLOW_DESTRUCTIVE

unset = off

Set to true to allow delete_data and setup_pre_commit (Git hook install).

HELA_GENOME_ALLOW_EXTERNAL_PATHS

unset = off

Set to true to let import_data/export_data use absolute or traversal paths outside the active project root. Off = paths must stay inside the workspace.

HELA_ENVELOPE

unset = off

Set to true to wrap tool results in the canonical HeLaResult envelope (ok/summary/data/artifacts/provenance/warnings/sideEffects/execution). Off = byte-identical legacy output. Run/step ids propagate from HELA_RUN_ID/HELA_STEP_ID.

MCP clients launch servers with their own working directory, which is often your home folder rather than the project you are editing. In that situation Git detection cannot find the project and every session writes to the shared fallback database. Two ways to fix this:

  1. Set GUARDIAN_PROJECT_ROOT in the project's MCP configuration (see the client examples below).

  2. Call the set_project_root tool with the absolute project path at session start — no configuration edits needed. The switch applies to the running server; set the environment variable if you want it to apply automatically to every future session.

Optional Runtime Services

Redis is optional and is never contacted during startup. Configure it only when cache tools are needed:

{
  "env": {
    "REDIS_URL": "redis://localhost:6379/0"
  }
}

Trivy is discovered from PATH when scan_container_image is called. Missing Redis or Trivy affects only its associated tools; memory, database, guidance, session, Git, wall, and project-secret tools remain available.

The companion catalog reports available, optional, or unavailable for each runtime capability. The server uses stdio transport and does not expose an HTTP listener.

For Cursor IDE

Add this server to your Cursor MCP configuration (~/.cursor/mcp.json). Replace the GUARDIAN_PROJECT_ROOT value with the project this configuration belongs to:

{
  "mcpServers": {
    "project-guardian": {
      "command": "node",
      "args": ["/path/to/project-mcp/dist/index.js"],
      "env": {
        "GUARDIAN_PROJECT_ROOT": "/path/to/your/project"
      }
    }
  }
}

For Claude Desktop

Add this server to your Claude Desktop configuration (claude_desktop_config.json), following the same pattern:

{
  "mcpServers": {
    "project-guardian": {
      "command": "node",
      "args": ["/path/to/project-mcp/dist/index.js"],
      "env": {
        "GUARDIAN_PROJECT_ROOT": "/path/to/your/project"
      }
    }
  }
}

Project Structure

project-mcp/
├── src/
│   ├── index.ts              # Main entry point
│   ├── server.ts             # MCP server orchestrator
│   ├── memory-manager.ts     # Knowledge graph and FTS5 RAG semantic search
│   ├── sqlite-manager.ts     # Database operations and connection management
│   ├── import-export.ts      # CSV/JSON data import and export functionality
│   ├── obsidian-vault.ts     # Optional Obsidian vault mirror (human view)
│   ├── types.ts              # TypeScript type definitions and schemas
│   ├── handlers/
│   │   └── request-handlers.ts # Central tool execution dispatcher
│   ├── tools/
│   │   ├── tool-registry.ts     # Tool definitions and listing
│   │   ├── database-tools.ts    # Database operation tool schemas
│   │   ├── memory-tools.ts      # Memory management tool schemas
│   │   ├── guidance-tools.ts    # Guidance tool schema
│   │   └── runtime-tools.ts     # Companion runtime tool schemas
│   ├── runtime/
│   │   ├── path-guard.ts        # Workspace path containment
│   │   └── runtime-capabilities.ts # Native companion implementations
│   ├── resources/
│   │   ├── resource-registry.ts  # Resource definitions and handlers
│   │   ├── resource-definitions.ts # Static resource metadata
│   │   ├── resource-handlers.ts   # Dynamic resource content generation
│   │   └── companion-catalog.ts   # Companion capability health
│   └── prompts/
│       ├── prompt-registry.ts       # Prompt definitions and handlers
│       ├── prompt-definitions.ts    # Static prompt metadata
│       ├── prompt-handlers.ts       # Dynamic prompt content generation
│       └── behavioral-protocol.ts   # Shared Behavioral Protocol system prompt
├── (no ui/ directory)        # Web UI dashboard removed; Obsidian vault is the human view
│   └── vite.config.ts        # Vite build configuration
├── __tests__/                # Comprehensive test suite
│   ├── tool-registry.test.ts
│   ├── resource-registry.test.ts
│   ├── prompt-registry.test.ts
│   ├── request-handlers.test.ts
│   ├── runtime-capabilities.test.ts
│   ├── import-export.test.ts
│   ├── sqlite-manager.test.ts
│   └── bug-fixes.test.ts
├── skills/                   # Six distributable guardian-* AgentSkills
├── dist/                     # Ignored production build output
├── memory.db                 # Ignored local SQLite state, created on first run
├── package.json              # Project dependencies and scripts
├── package.prod.json         # Production-only dependencies for smaller bundle
├── tsconfig.json            # TypeScript configuration
├── jest.config.js           # Test configuration
└── README.md                # This documentation

Key Components

  • server.ts: MCP server lifecycle, transport, handlers, and shutdown coordination

  • handlers/request-handlers.ts: Central dispatcher routing tool calls to appropriate managers

  • tools/: Tool definition and registration system (29 canonical tools; legacy action aliases are not additional advertised tools)

    • tool-registry.ts: Lists all available tools (7 DB + 12 memory + 1 guidance + 12 runtime + 2 session bridge + 4 UI/stream)

    • database-tools.ts: Database operation schemas (7 tools)

    • memory-tools.ts: Memory management schemas (12 tools)

    • session-bridge-tools.ts: Session bridge ingestion schemas (2 tools)

    • guidance-tools.ts: Autonomous guidance tool schema (1 tool)

    • runtime-tools.ts: Typed companion capability schemas (12 tools)

  • hooks/: Zero-touch lifecycle hooks (session-start.mjs, post-tool-run.mjs)

  • runtime/: Workspace guards and companion runtime implementations

  • resources/: Resource management system (11 resources total)

    • resource-registry.ts: Resource listing and content serving

    • resource-definitions.ts: Static resource metadata

    • resource-handlers.ts: Dynamic content generation

  • prompts/: Prompt management system (27 prompts total)

    • prompt-registry.ts: Prompt listing and content serving

    • prompt-definitions.ts: Static prompt metadata

    • prompt-handlers.ts: Dynamic prompt generation with context

    • behavioral-protocol.ts: Centralized Behavioral Protocol system message used by all prompts

  • memory-manager.ts: Knowledge graph operations for entities, relationships, and observations

  • sqlite-manager.ts: Database abstraction with bounded connection caching and schema management

  • import-export.ts: CSV, JSON, and SQL data transfer utilities

  • types.ts: Zod schemas for input validation and TypeScript type safety

  • skills/: Agent-side workflows, scripts, references, and assets for the six companion packages

Local State

memory.db and its memory.db-* sidecars are runtime state and are ignored by Git. Each project keeps its own database at its resolved project root (see Environment Variables); projects outside any Git repository without an explicit root share the fallback database under $XDG_DATA_HOME/project-guardian. In addition, every memory write mirrors into the central database at ~/memory/memory.db, which is a merge across projects: deleting an entity in one project does not remove it from the central copy, so treat the central database as a searchable aggregate rather than a per-project backup. Daily snapshots live in ~/memory/backup/. A clone starts with no project memory; the server creates the database and schema locally on first run. Back up or export memory explicitly when it must move between machines. Never commit the database because observations can contain private project context.

The database tools (execute_sql, query_data, insert_data, update_data, delete_data, import_data, export_data) accept a database selector: project (default) targets the active project database, central targets the aggregate.

Development

  1. Clone the repository:

git clone https://github.com/Hela-Ecosystem/project-mcp.git
cd project-mcp
  1. Install dependencies:

npm install
  1. Build the project: For active development (with file watching):

npm run dev

For a standard build:

npm run build

For a production-optimized build:

npm run build:prod
  1. Run tests:

npm test
  1. Start the server:

npm start

License

GPL-3.0-or-later License - see LICENSE file for details.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      A persistent memory server for AI agents that stores structured notes in a local SQLite database with full-text search and graph-based relationships. It features 32 specialized tools for managing long-term context, including version history, automated TTL expiration, and complex filtering.
      12 npm
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides Claude with a persistent local memory and structured knowledge graph to track project states, tasks, and historical decisions across different chat sessions. It enables users to recall information using keyword relevance, time-travel queries, and dependency analysis for complex project management.
      MIT
    • A
      license
      B
      quality
      D
      maintenance
      Ultra-lean memory system for AI coding tools that stores project knowledge locally with SQLite and enables AI to remember your project across sessions.
      12
      85 npm
      37
      MIT