vpursuit/swipl-mcp-server
If you are the rightful owner of swipl-mcp-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.
The SWI-Prolog MCP Server is a Model Context Protocol server designed to integrate SWI-Prolog with Large Language Model applications, enabling direct interaction and tool calling capabilities.
SWI-Prolog MCP Server
Package | License | Node |
---|---|---|
@vpursuit/swipl-mcp-server | BSD-3-Clause | ≥20.0.0 |
An MCP server that lets tools-enabled LLMs work directly with SWI‑Prolog. It supports loading Prolog files, adding/removing facts and rules, listing symbols, and running queries with two modes: deterministic pagination and true engine backtracking.
Table of Contents
- Requirements
- Installation
- Configuration
- Features
- Examples
- Architecture
- Security
- Troubleshooting
- Development
- Contributing
- Documentation
- License
Requirements
- Node.js ≥ 20.0.0
- SWI‑Prolog installed and available in PATH
Installation
Claude Code CLI
claude mcp add swipl-mcp-server npx @vpursuit/swipl-mcp-server
Claude Desktop
{
"mcpServers": {
"swipl": {
"command": "npx",
"args": ["@vpursuit/swipl-mcp-server"]
}
}
}
- MacOS:
~/Library/Application Support/Claude/claude_desktop_config.json
- Windows:
%APPDATA%/Claude/claude_desktop_config.json
Cline (VS Code Extension)
{
"mcpServers": {
"swipl-mcp-server": {
"autoApprove": [],
"disabled": false,
"timeout": 60,
"type": "stdio",
"command": "npx",
"args": ["@vpursuit/swipl-mcp-server"]
}
}
}
Configure via Cline's MCP settings in VS Code.
Codex
[mcp_servers.swipl-mcp-server]
transport = "stdio"
enabled = true
command = "npx"
args = ["@vpursuit/swipl-mcp-server"]
Add to ~/.codex/config.toml
MCP Inspector (for testing)
npx @modelcontextprotocol/inspector --transport stdio npx @vpursuit/swipl-mcp-server
... and many others may also work
Development Setup
If you cloned the repo, you may use this configuration. Note: change
{
"mcpServers": {
"swipl": {
"command": "node",
"args": ["<path to your development directory>/swipl-mcp-server/build/index.js"],
"env": {
"SWI_MCP_READY_TIMEOUT_MS": "10000",
"SWI_MCP_QUERY_TIMEOUT_MS": "120000",
"MCP_LOG_LEVEL": "debug",
"DEBUG": "swipl-mcp-server"
}
}
}
}
Configuration
Environment Variables
Configure timeouts, logging, and behavior via environment variables:
SWI_MCP_READY_TIMEOUT_MS
: server startup timeout (ms), default 5000SWI_MCP_QUERY_TIMEOUT_MS
: query execution timeout (ms), default 30000MCP_LOG_LEVEL
:debug
|info
|warn
|error
|silent
(defaultwarn
)DEBUG
: enable debug logs, set toswipl-mcp-server
SWI_MCP_TRACE
: optional low-level trace of child I/O and protocolSWI_MCP_PROLOG_PATH
: override Prolog server script path
State & Lifecycle
- Transport:
stdio
. The MCP client owns the connection lifecycle. - Shutdown: the server exits on
SIGINT
/SIGTERM
or when the client closes stdio. On stdio close, a small grace (~25ms) allows final responses to flush before exit. - Stateful per connection: asserted facts/rules live in memory for the lifetime of the MCP connection (one Node process and one SWI‑Prolog child). When the client disconnects and the server exits, in‑memory state is reset on next start.
- Client guidance: keep a single stdio connection open for workflows that depend on shared state across multiple tool calls; avoid closing stdin immediately after a request.
- Durability (optional): if persistent Knowledge Base is desired across restarts, use
knowledge_base_dump
to save to~/.swipl-mcp-server/
andknowledge_base_load
(orknowledge_base_assert_many
) to restore on startup. See docs/lifecycle.md for patterns.
Features
MCP Prompts
Prompts guide AI assistants to help you with Prolog programming, knowledge base building and query optimization.
How it works:
- You select a prompt (via
/swipl
command in Claude Code CLI) - The prompt guides the AI assistant on how to approach your Prolog task
- The AI assistant helps you with expert knowledge and step-by-step guidance
Note: Other AI assistants may access and use these prompts differently depending on their MCP implementation.
In Claude Code CLI, these prompts are available as slash commands. Simply type /swipl
to see all available commands:
Available prompts:
prolog_init_expert
- Initialize expert Prolog assistance mode with optional task focusprolog_quick_reference
- Get comprehensive server overview and capabilitiesprolog_analyze_knowledge_base
- Analyze current knowledge base state and structureprolog_knowledge_base_builder
- Build domain-specific knowledge bases with guided constructionprolog_query_optimizer
- Optimize Prolog queries for performance and efficiency
MCP Resources
Dynamic and static resources for knowledge base access:
prolog://knowledge_base/predicates
- List all predicates in the knowledge baseprolog://knowledge_base/dump
- Export complete knowledge base as Prolog clausesreference://help
- Usage guidelines and server tipsreference://license
- BSD-3-Clause license textreference://capabilities
- Machine-readable server capabilities (JSON)
Tools
- Core:
help
,license
,capabilities
- Knowledge base:
knowledge_base_load
,knowledge_base_assert
,knowledge_base_assert_many
,knowledge_base_retract
,knowledge_base_retract_many
,knowledge_base_clear
,knowledge_base_dump
- Query:
query_start
,query_startEngine
,query_next
,query_close
- Symbols:
symbols_list
Available Predicates
All standard SWI-Prolog predicates are available (lists, arithmetic, meta-predicates, etc.).
Note: CLP(FD) (library(clpfd)
) is not available for security reasons. Use standard Prolog alternatives:
between/3
instead ofX in 1..10
is/2
instead of#=
permutation/2
for generating unique values- Generate-and-test pattern instead of constraint propagation
Examples
Loading and Querying Knowledge Base
Load a Prolog file (files must be in ~/.swipl-mcp-server/
):
knowledge_base_load { "filename": "~/.swipl-mcp-server/family.pl" }
Start a query and iterate through solutions:
query_start { "query": "parent(X, mary)" }
query_next() // Get first solution
query_next() // Get next solution
query_close() // Close when done
Engine Mode (True Backtracking)
For queries requiring all solutions or complex backtracking:
query_startEngine { "query": "member(X, [1,2,3])" }
query_next() // X = 1
query_next() // X = 2
query_next() // X = 3
query_next() // No more solutions
query_close()
Database Operations
Add facts:
// Single fact
knowledge_base_assert { "fact": "parent(john, mary)" }
// Multiple facts
knowledge_base_assert_many {
"facts": ["parent(john, mary)", "parent(mary, alice)"]
}
Remove facts:
// Single fact
knowledge_base_retract { "fact": "parent(john, mary)" }
// Multiple facts
knowledge_base_retract_many {
"facts": ["parent(john, mary)", "parent(mary, alice)"]
}
// Clear all user facts
knowledge_base_clear {}
More Examples
See for comprehensive examples including arithmetic, list operations, collections, and string/atom helpers.
Architecture
- Single persistent SWI‑Prolog process with two query modes (standard via
call_nth/2
, engine via SWI engines) - Term-based wire protocol: Node wraps requests as
cmd(ID, Term)
, replies asid(ID, Reply)
; back‑compatible with bare terms - Enhanced security model with file path restrictions, library(sandbox) validation, and dangerous predicate blocking
Details: see docs/architecture.md.
Session State Machine
The server maintains a session state machine to coordinate query and engine sessions. Key points:
- Exactly one session type can be active at a time (query or engine)
- The
*_completed
states keep context so that subsequentnext
calls respond with "no more solutions" until explicitly closed - Transient
closing_*
states serialize shutdown before new sessions begin - Invalid transitions are logged when
SWI_MCP_TRACE=1
For the detailed state transition diagram, see .
Security
The server implements multiple security layers to protect your system:
File Path Restrictions
- Allowed Directory: Files can only be loaded from
~/.swipl-mcp-server/
- Blocked Directories: System directories (
/etc
,/usr
,/bin
,/var
, etc.) are automatically blocked - Example:
→
knowledge_base_load { "filename": "/etc/passwd" }
Security Error: Access to system directories is blocked
Dangerous Predicate Detection
- Pre-execution Blocking: Dangerous operations are caught before execution
- Blocked Predicates:
shell()
,system()
,call()
,assert()
,halt()
, etc. - Example:
→
knowledge_base_assert { "fact": "malware :- shell('rm -rf /')" }
Security Error: Operation blocked - contains dangerous predicate 'shell'
Additional Protections
- Library(sandbox) validation for built-in predicates
- Timeout protection against infinite loops
- Module isolation in dedicated
knowledge_base
namespace
See for complete security documentation.
Troubleshooting
- "Prolog not found": ensure
swipl --version
works; SWI‑Prolog must be in PATH - Startup timeout: increase
SWI_MCP_READY_TIMEOUT_MS
- Query timeout: increase
SWI_MCP_QUERY_TIMEOUT_MS
- Session conflicts: close current session before starting a different mode
Security Error: ...
: file access blocked or dangerous predicates detected; see Security- Custom script path: set
SWI_MCP_PROLOG_PATH
- Query sessions: after exhausting solutions,
query_next
returns "No more solutions available" until explicitly closed
Development
- Install deps:
npm install
- Build:
npm run build
- Run dev server:
npm run server
- Tests:
npm test
(see CONTRIBUTING.md for details)
Publishing and release workflows are documented in docs/deployment.md.
Contributing
See CONTRIBUTING.md for local setup, workflow, and the PR checklist.
For security practices, reporting, and hardening guidance, see SECURITY.md.
Documentation
- — Complete setup for all MCP clients
- — Detailed prompts, resources, and tools documentation
- — Copy-paste usage examples
- — Components, modes, and wire protocol
- — Server lifecycle, state, and persistence patterns
- — Release, packaging, and install from source
License
BSD‑3‑Clause. See LICENSE for details.