MAXential Thinking MCP

A structured, persistent reasoning workspace for AI. Provides 20 tools for building thought chains, exploring alternatives through branching, revising earlier thinking, searching reasoning history, and persisting sessions across context window resets and server restarts.

AI's built-in reasoning is ephemeral — when context windows fill, thinking gets compressed or lost. Complex problems need exploration of multiple approaches, backtracking when paths fail, and the ability to resume where you left off. MAXential externalizes reasoning into a workspace that persists, branches, and survives.

Forked fromAnthropic's sequential-thinking MCP server, which provided a single tool with 9 parameters. Its schema included branching parameters (branch_from_thought,branch_id) but had no tools to create, switch, or manage branches — the parameters were effectively inert.

sqlite3 .maxential/thinking.db "SELECT name, status, datetime(created_at/1000, 'unixepoch', 'localtime') as created FROM sessions ORDER BY updated_at DESC LIMIT 10;"

{ "mcpServers": { "maxential-thinking": { "command": "npx", "args": ["-y", "@bam-devcrew/maxential-thinking-mcp"] } } }

git clone https://github.com/BAM-DevCrew/MAXential-Thinking-MCP.git cd MAXential-Thinking-MCP npm install npm run build

{ "mcpServers": { "maxential-thinking": { "command": "node", "args": ["/path/to/MAXential-Thinking-MCP/dist/src/index.js"] } } }

Session data is stored in SQLite. By default, the database is created at.maxential/thinking.dbin the working directory.

{ "mcpServers": { "maxential-thinking": { "command": "npx", "args": ["-y", "@bam-devcrew/maxential-thinking-mcp"], "env": { "MAXENTIAL_DB_PATH": "/path/to/your/thinking.db" } } } }

If SQLite initialization fails (permissions, native module issues), the server falls back to in-memory mode automatically — it never crashes.

Add.maxential/to your project's.gitignoreto keep session data out of version control.

{ "mcpServers": { "maxential-thinking": { "command": "npx", "args": ["-y", "@bam-devcrew/maxential-thinking-mcp"], "env": { "MAXENTIAL_LOG_FILE": "/path/to/error.log" } } } }

You don't call these tools directly — you ask your AI to use MAXential thinking, and it calls the tools as part of its reasoning. Here are examples of what that looks like in practice.

use maxential thinking for this - should we use REST or GraphQL for the new API?

AI builds a thought chain analyzing the question, branches to explore each approach separately, adds thoughts with tradeoffs, merges the findings, and reaches a conclusion. The entire reasoning process is numbered, structured, and persisted.

think through the auth redesign using maxential - I want to see branches for JWT, session tokens, and OAuth

AI creates three branches, reasons through each approach independently, then merges the insights back to compare. You can ask it to switch between branches, close dead ends, or dig deeper into a specific path.

load up that session where you analyzed our database optimization options

AI browses saved sessions, finds the match, restores it with all thoughts, branches, and tags intact, and continues reasoning from where it left off — even across different conversations.

search your maxential thinking history for anything tagged as a decision

AI searches across the session's tagged thoughts and returns the results. You can also ask it to export the full chain as markdown, or generate a diagram of the thought structure.

give me a summary of that session - just the key findings, keep it short

AI generates a compressed summary of the session's conclusions, tagged highlights, and branch results — useful for loading context without replaying the full chain.

npm install # Install dependencies npm run build # Build TypeScript npm run watch # Watch mode npm test # Run tests with coverage npm run test:unit # Unit tests only npm run test:integration # Integration tests only

Issues and PRs welcome atgithub.com/BAM-DevCrew/MAXential-Thinking-MCP

Score CLAUDE.md/AGENTS.md, validate clarx-manifest.json, generate CI workflows, and pull repo scan findings into Cursor, Claude, or Grok.