zpm

A high-performance MCP (Model Context Protocol) server written in Zig, designed to bridge Large Language Models with a Prolog inference engine for deterministic logical reasoning.

Features

• CLI entrypoint with init, serve, upgrade, and all 29 MCP tools exposed as subcommands (e.g. zpm remember-fact, zpm query-logic)
• Self-upgrade via zpm upgrade with SHA256 verification, atomic install, and --channel stable|dev selection
• Per-project .zpm/ directory for isolated configuration and persistence
• MCP protocol version 2025-11-25 over STDIO transport
• Tool registration and discovery via tools/list
• Echo tool for health-check and smoke testing
• Prolog inference engine with Trealla C API integration
• Knowledge management tools: assert facts and define rules via MCP
• Exploration tools: query Prolog goals and trace transitive dependencies via MCP
• Supervision tools: verify knowledge base consistency and explain proof chains via MCP
• Knowledge schema discovery: inspect file-loaded static facts and rules alongside asserted predicates, with clause counts and types, in each selected memory
• Delete tools: retract individual facts or clear entire categories from the knowledge base
• Update tools: atomically replace individual facts or upsert facts with pattern matching
• Truth Maintenance System: manage assumptions and automatically propagate belief changes via MCP
• Knowledge base persistence: durable storage via write-ahead journal and fresh-engine-validated snapshots with automatic recovery
• Named memory segments for domain-isolated knowledge bases with independent WAL/snapshot persistence, read-only mount mode, project and global scope, cross-memory queries via Prolog module qualification, and discoverable memory selectors in MCP schemas and CLI help
• Persistent mount manifest (.zpm/mounts.json) for durable memory configuration across CLI invocations
• Documentation site with Hugo/Thulite (Doks) and GitHub Pages auto-deployment
• Zero external runtime dependencies (statically linked, including Prolog library)

Quick Start

Prerequisites

• Zig >= 0.16.0
• C compiler (gcc or clang)

Install

Install the latest pre-built binary with a single command:

curl -fsSL https://raw.githubusercontent.com/awf-project/zpm/main/scripts/install.sh | sh

Supported platforms: Linux (x86_64, arm64), macOS (x86_64, arm64).

Build from Source

# Build the server binary
make build

# Run unit tests
make test

# Run functional (end-to-end) tests
make functional-test

The built binary is located at zig-out/bin/zpm.

CLI Usage

# Display help (zpm with no args also prints help, but exits 1)
zpm --help

# Show version (use the subcommand; --version/-v fall through to help)
zpm version

# Initialize a project directory
zpm init

# Start the MCP server (STDIO transport)
zpm serve

# Upgrade to the latest release (verifies SHA256, atomic replace)
zpm upgrade
zpm upgrade --channel dev --dry-run

# Invoke MCP tools directly from the shell
zpm remember-fact --fact "task_status(f017, done)"
zpm query-logic --goal "task_status(X, done)" --format json
zpm save-snapshot --name "before-deploy"

# Manage memory segments
zpm memory create --name feature_auth
zpm memory create --name shared_profiles --scope global
zpm memory list
zpm remember-fact --fact "task_done(login)" --memory feature_auth
zpm query-logic --goal "task_done(X)" --memory feature_auth
zpm get-knowledge-schema --memory feature_auth
zpm memory mount --name feature_auth --mode ro

The optional --memory flag selects a mounted segment. Omitting it keeps unqualified reads in the default memory. Memory-aware MCP tools advertise the optional string memory argument in tools/list, so clients can discover it before calling a tool. Use zpm get-knowledge-schema --memory <name> to inspect predicates in a named segment; see Schema Discovery for an example.

See the CLI Reference for the full list of tool subcommands and flags.

Connect via MCP Client

Add zpm to your MCP client configuration. For example, in Claude Code's settings.json:

{
  "mcpServers": {
    "zpm": {
      "command": "zpm",
      "args": ["serve"]
    }
  }
}

Commands

Command	Description
make build	Build the server binary
make check	Run all checks (lint + test + e2e)
make clean	Remove build artifacts
make fmt	Format source code
make functional-test	Run end-to-end MCP protocol tests
make lint	Check formatting
make test	Run unit tests (inline Zig tests)
make upgrade-test	Run upgrade end-to-end tests

MCP Tools

Tool	Description	Arguments
assume_fact	Assert a fact under a named assumption with automatic justification tracking	assumption (string, required), fact (string, required), memory (string, optional)
clear_context	Clear all facts from the knowledge base matching a pattern	category (string, required), memory (string, optional)
create_memory	Create a named memory segment backed by a dedicated Prolog module	name (string, required), scope (string, optional)
define_rule	Assert a Prolog rule into the knowledge base	head (string, required), body (string, required), memory (string, optional)
echo	Returns the provided message (health-check)	message (string, required)
explain_why	Trace proof tree for a fact and return structured deduction chain	fact (string, required), max_depth (integer, optional), memory (string, optional)
find_predicate_references	Locate all usage sites of a Prolog predicate in the knowledge base	functor (string, required), arity (integer, optional), memory (string, optional), include_cross_memory_refs (boolean, optional)
forget_fact	Remove a single fact from the knowledge base	fact (string, required), memory (string, optional)
get_belief_status	Query whether a belief is currently supported and which assumptions justify it	fact (string, required), memory (string, optional)
get_justification	Query all facts supported by a named assumption	assumption (string, required), memory (string, optional)
get_kb_overview	Single-call snapshot of the default knowledge base: static and asserted user predicates with sample clauses, assumptions, snapshots, and persistence health	sample_size (integer, optional)
get_knowledge_schema	List file-loaded static and asserted user predicates in the selected memory with arity, type (fact/rule/both), and clause count	memory (string, optional)
get_persistence_status	Query the persistence layer status including journal size, last snapshot, and operational mode	(no required arguments)
list_assumptions	List all currently active assumptions and their associated facts	memory (string, optional)
list_memories	List all discoverable memories with scope, mount status, and mode	(no required arguments)
list_snapshots	List all available knowledge base snapshots with metadata	memory (string, optional)
mount_memory	Mount a named memory segment for use, optionally in read-only mode	name (string, required), mode (string, optional)
query_logic	Execute a Prolog goal and return all variable bindings as JSON	goal (string, required), memory (string, optional)
remember_fact	Assert a Prolog fact into the knowledge base	fact (string, required), memory (string, optional)
rename_predicate	Atomically rename a Prolog functor across facts, rules, and TMS justifications	old_functor (string, required), new_functor (string, required), arity (integer, optional), memory (string, optional), dry_run (boolean, optional), propagate_cross_memory_refs (boolean, optional)
restore_snapshot	Restore knowledge base from a named snapshot and replay subsequent journal entries	name (string, required), memory (string, optional)
retract_assumption	Retract a named assumption and propagate its removal through the knowledge base	assumption (string, required), memory (string, optional)
retract_assumptions	Retract all assumptions matching a glob-style pattern	pattern (string, required), memory (string, optional)
save_snapshot	Create a named point-in-time snapshot of the knowledge base	name (string, required), memory (string, optional)
trace_dependency	Trace transitive dependencies from a start node using path/2 rules	start_node (string, required), memory (string, optional)
unmount_memory	Unmount a named memory, flushing WAL and freeing resources	name (string, required)
update_fact	Atomically replace an existing fact (retract old, assert new)	old_fact (string, required), new_fact (string, required), memory (string, optional)
upsert_fact	Replace a fact matching functor+first arg, or insert if not found	fact (string, required), memory (string, optional)
verify_consistency	Check knowledge base for integrity violations	scope (string, optional), memory (string, optional)

Documentation

Browse the documentation online at https://awf-project.ai/zpm/ or in the docs/ directory:

• Project Brief -- Vision and objectives
• Getting Started -- Build, install, and first steps
• Reference -- MCP tools and protocol details
• Snapshot recovery -- Repair snapshots created by affected versions
• ADR -- Architecture Decision Records

To preview the documentation site locally:

cd site && npm ci && npm run dev

License

See LICENSE.