Adds persistent reasoning and memory systems to Claude through six engines: symbolic (forward/backward chaining), knowledge graph traversal, case-based retrieval, hypothesis testing, and analogical reasoning. Exposes 11 MCP tools including reason_hybrid (auto-selects best approach), remember/recall for storage, and learn for problem-solution pairs. State persists as JSON in ~/.bolor-brain/ so solved bugs and decisions compound over sessions. Works standalone or pairs with NSAF MCP for strategy evolution. Designed to complement Claude Code's execution layer with pure cognitive operations. Best for debugging complex issues, making technical decisions with historical context, or building up domain knowledge that survives across conversations.
Pure intelligence for Claude Code. Reasoning, memory, and learning -- nothing else.
Bolor Brain is an MCP server that gives Claude Code a brain: structured reasoning, persistent memory, and learning from experience.
User --> Claude Code (Gateway + Executor)
|
+----+----+
Bolor Brain NSAF
(MCP) (MCP)
THINK EVOLVE
Bolor Brain does NOT execute anything. No file ops, no scheduling, no autonomous loop. Claude Code already does all of that. Bolor Brain only thinks.
git clone https://github.com/photoxpedia/bolor-brain-mcp.git
cd bolor-brain-mcp
pip install -e .
Add to ~/.claude/mcp-config.json:
{
"mcpServers": {
"bolor-brain": {
"command": "python",
"args": ["-m", "mcp_server"],
"cwd": "/path/to/bolor-brain-mcp"
}
}
}
/reason Why is Python popular for data science?
/debug API returns 500 errors under load
/decide PostgreSQL or MongoDB for our app?
/learn-from We fixed the memory leak by increasing connection pool
| Tool | What It Does |
|---|---|
reason_hybrid | Auto-selects best reasoning approach for any query |
reason_symbolic | Forward/backward chaining with facts and rules |
reason_knowledge_graph | Graph traversal, path finding, relationship exploration |
reason_case_based | Find similar past problems and their solutions |
reason_hypothesis | Generate and test hypotheses from observations |
reason_analogical | Cross-domain pattern transfer (atom ~ solar system) |
| Tool | What It Does |
|---|---|
remember | Store a case, fact, node, or edge |
recall | Retrieve matching cases or facts |
learn | Store problem/solution/outcome (shortcut for remember) |
forget | Delete a case or fact by ID |
| Tool | What It Does |
|---|---|
brain_stats | Cases, facts, nodes, edges count |
| Skill | When To Use |
|---|---|
/reason | Deep analysis of any complex problem |
/debug | Systematic bug hunting with hypothesis testing |
/decide | Evidence-based technical decisions |
/learn-from | Store experiences for future use |
/nsaf | NSAF evolution integration (requires NSAF MCP) |
/orchestrate | Meta-orchestration combining Bolor Brain + NSAF |
Brain state persists to ~/.bolor-brain/ as JSON:
~/.bolor-brain/
cases.json # Problem -> solution -> outcome
facts.json # Symbolic reasoning facts
knowledge.json # Knowledge graph (nodes + edges)
Knowledge compounds over time. Solve a bug once, recall the solution instantly next time.
Add NSAF to get evolution capabilities:
{
"mcpServers": {
"bolor-brain": {
"command": "python",
"args": ["-m", "mcp_server"],
"cwd": "/path/to/bolor-brain-mcp"
},
"nsaf": {
"command": "python3",
"args": ["nsaf_mcp_server.py"],
"cwd": "/path/to/nsaf",
"env": { "PYTHONPATH": "/path/to/nsaf" }
}
}
}
Together: Bolor Brain reasons about WHAT to do. NSAF evolves HOW to do it better. Claude Code executes.
See skills/nsaf.md and skills/orchestrate.md for combined workflows.
pytest tests/ -v
# 376 tests
mcp_server.py # MCP server (11 tools)
persistence.py # JSON persistence to ~/.bolor-brain/
modules/
config.py # Configuration
reasoning_engines/
symbolic_reasoner.py # Forward/backward chaining
knowledge_graph.py # Graph-based knowledge
case_based_reasoner.py # 4R cycle (retrieve, reuse, revise, retain)
hypothesis_engine.py # Hypothesis generation and testing
analogical_reasoner.py # Cross-domain pattern transfer
hybrid_reasoner.py # Orchestrates all 5 engines
skills/ # Claude Code skills
reason.md, debug.md, decide.md, learn-from.md, nsaf.md, orchestrate.md
tests/ # 376 tests
AGENT_GUARDRAILS.md # Production safety guidelines
Bolorerdene Bundgaa
MIT -- see LICENSE
BRAIN_USE_SQLITEEnable SQLite storage (default: true)
BRAIN_LOG_LEVELLog level (default: INFO)
io.github.ericm1018/skillfm-llm-cost-optimizer-openai-anthropic-usage
io.github.mikerawsonnz/llm-orchestration-agent
io.github.mikerawsonnz/authenticated-llm-agent
labforgedev/copilot-memory-mcp
csoai-org/agent-prompt-injection-firewall-mcp
io.github.mikerawsonnz/authenticated-multi-llm-agent