Skip to main content
Glama
NikkeTryHard

Tree of Thoughts MCP Server

by NikkeTryHard
OverviewSchemaRelated ServersScoreDiscussions
TypeScript
Local

Tree of Thoughts MCP Server

A Model Context Protocol (MCP) server that enables structured multi-path investigations using a Tree of Thoughts approach. Designed for AI agents to explore complex problems systematically with parallel research paths.

Features

  • Single Root Paradigm - Start with one root, branch wide at R2

  • 5-State System - EXPLORE, FOUND, VERIFY, EXHAUST, DEAD states enforce thorough investigation

  • Anti-Gaming Measures - Timing checks and agentId tracking prevent shortcut-taking

  • Reference Extraction - Automatically collects URLs and file paths from findings

  • DOT Graph Output - Visualize investigation tree structure

Related MCP server: LLM Responses MCP Server

Installation

# Clone the repository
git clone https://github.com/NikkeTryHard/tree-of-thoughts-mcp.git
cd tree-of-thoughts-mcp

# Install dependencies
bun install

# Build
bun run build

Configuration

Add to your Claude Code MCP settings (~/.claude/settings.json):

{
  "mcpServers": {
    "tree-of-thoughts": {
      "command": "node",
      "args": ["/path/to/tree-of-thoughts-mcp/dist/index.js"]
    }
  }
}

State Machine

State

Meaning

Children Required

Terminal

Min Round

Valid Child States

EXPLORE

Dig deeper

2+ (R1-R2), 1+ (R3+)

No

R1

Any

FOUND

Provisional solution

1+ any

No

R4

EXPLORE, FOUND, VERIFY

VERIFY

Confirms FOUND

0

Yes

R4

-

EXHAUST

Exhausted path

1+ any

No

R4

EXPLORE, EXHAUST, DEAD

DEAD

Confirmed dead end

0

Yes

R4

-

Key insight: FOUND is NOT terminal. Every promising finding must be verified by at least one child before the investigation can end.

Key insight: EXHAUST is NOT terminal. Every exhausted path must be confirmed by at least one child.

Invalid Transitions (Semantic Conflicts):

From

Cannot Go To

Why

FOUND

EXHAUST, DEAD

Found something - can't be exhausted/dead

EXHAUST

FOUND, VERIFY

Exhausted path - can't suddenly find/verify

Workflow (Single Root Paradigm)

1. tot_start    → Get sessionId, begin investigation
2. tot_propose  → Declare single root R1.A
3. Spawn agent  → Execute research for R1.A
4. tot_commit   → Submit as EXPLORE
5. tot_propose  → Branch into 3-5 children at R2
6. Repeat       → Continue until R4+ for FOUND, R5+ to end
7. tot_end      → Finalize and get results with references

Node ID Format

Round 1: R1.A (single root, parent: null)
Round 2: R2.A1, R2.A2, R2.A3 (branch wide from R1.A)
Round 3: R3.A1a (parent: R2.A1)
Round 4: R4.A1a1 (parent: R3.A1a) - can use FOUND here
Round 5: R5.A1a1a (parent: R4.A1a1) - VERIFY node

Rules

  1. Single root R1.A - Then branch wide at R2 (5+ nodes recommended)

  2. Minimum 5 rounds - Cannot end before Round 5

  3. EXPLORE needs 2+ children at R1-R2, 1+ at R3+ - Relaxed requirements later

  4. R3 must be EXPLORE only - Forces deeper investigation before conclusions

  5. FOUND only at R4+ - Earlier rounds auto-convert to EXPLORE

  6. FOUND needs 1+ child (EXPLORE, FOUND, or VERIFY) - Cannot end until verified

  7. EXHAUST only at R4+ - Earlier rounds auto-convert to EXPLORE

  8. EXHAUST needs 1+ child (EXPLORE, EXHAUST, or DEAD) - Cannot end unconfirmed

  9. DEAD only at R4+ - R1-R3 converts to EXPLORE

Example

// Start investigation
tot_start({ query: "How to optimize database queries?" })

// Round 1: Single root
tot_propose({ sessionId, nodes: [
  { id: "R1.A", parent: null, title: "Query Optimization", plannedAction: "Analyze problem" }
]})
tot_commit({ sessionId, results: [
  { nodeId: "R1.A", state: "EXPLORE", findings: "Found multiple paths...", agentId: "abc-123" }
]})

// Round 2: Branch wide
tot_propose({ sessionId, nodes: [
  { id: "R2.A1", parent: "R1.A", title: "Index Analysis", plannedAction: "Analyze indexes" },
  { id: "R2.A2", parent: "R1.A", title: "Query Patterns", plannedAction: "Review patterns" },
  { id: "R2.A3", parent: "R1.A", title: "Schema Design", plannedAction: "Evaluate schema" }
]})

// Continue through R3, R4...

// Round 4+: Use FOUND
tot_commit({ sessionId, results: [
  { nodeId: "R4.A1a1", state: "FOUND", findings: "Solution: Add composite index...", agentId: "..." }
]})

// Round 5: Verify the finding
tot_propose({ sessionId, nodes: [
  { id: "R5.A1a1a", parent: "R4.A1a1", title: "Verify Index", plannedAction: "Test performance" }
]})
tot_commit({ sessionId, results: [
  { nodeId: "R5.A1a1a", state: "VERIFY", findings: "Confirmed: 10x improvement", agentId: "..." }
]})

// End when canEnd=true
tot_end({ sessionId })

Anti-Gaming Measures

The server includes protections against agents taking shortcuts:

Warning

Trigger

SUSPICIOUS

Commit within 10s of propose

MISSING_AGENT

No agentId provided

DEPTH_ENFORCED

FOUND before R4 (auto-converts)

Claude Code Skill

Skills are located in the tree-of-thoughts/ folder. Copy tree-of-thoughts/SKILL.md to ~/.claude/skills/tree-of-thoughts/SKILL.md to enable the /tree-of-thoughts command in Claude Code.

Development

# Run tests
bun test

# Build
bun run build

See CLAUDE.md for development guidelines.

License

MIT

This server cannot be installed

F
license - not found
-
quality - not tested
C
maintenance

How are these scores calculated?

Maintenance

–Maintainers
–Response time
–Release cycle
–Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/NikkeTryHard/tree-of-thoughts-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server