MCP Tools Reference

Complete reference for all tools exposed by the ChorumAI MCP server.

Why This Matters

Understanding what each tool does helps you get the most out of MCP integration—and helps you debug when things don’t work as expected.

Read Tools

These tools allow AI agents to read from your memory without requiring approval.

`chorum_list_projects`

List all projects accessible by the authenticated user.

Input Schema:

{
  // No required parameters
}

Output:

{
  projects: {
    id: string;
    name: string;
    lastActivity: string;  // ISO timestamp
  }[];
}

Example Response:

{
  "projects": [
    {
      "id": "proj_abc123",
      "name": "ChorumAI",
      "lastActivity": "2026-01-31T14:23:00Z"
    },
    {
      "id": "proj_def456",
      "name": "Marketing Site",
      "lastActivity": "2026-01-30T09:15:00Z"
    }
  ]
}

`chorum_query_memory`

Semantic search through project memory. Returns patterns, decisions, invariants, and facts ranked by relevance.

Input Schema:

{
  projectId: string;           // Required: which project's memory
  query: string;               // Natural language query
  types?: LearningType[];      // Optional filter: 'pattern' | 'antipattern' | 'decision' | 'invariant'
  maxTokens?: number;          // Budget limit (default: 2000, max: 8000)
  includeContext?: boolean;    // Include surrounding context (default: true)
}

Output:

{
  items: {
    id: string;
    type: LearningType;
    content: string;
    context?: string;
    relevanceScore: number;    // 0.0 to 1.0
    createdAt: string;
  }[];
  tokenCount: number;
  projectName: string;
}

Example Usage:

Agent: "What error handling patterns does this project use?"

Example Response:

{
  "items": [
    {
      "id": "learn_789",
      "type": "pattern",
      "content": "All API errors return structured JSON with code, message, and details fields",
      "relevanceScore": 0.87,
      "createdAt": "2026-01-15T10:00:00Z"
    },
    {
      "id": "learn_790",
      "type": "invariant",
      "content": "Never expose stack traces in production error responses",
      "relevanceScore": 0.82,
      "createdAt": "2026-01-10T16:30:00Z"
    }
  ],
  "tokenCount": 142,
  "projectName": "ChorumAI"
}

`chorum_get_invariants`

Get all active invariants for a project. Invariants are rules that must never be broken.

Input Schema:

{
  projectId: string;
}

Output:

{
  invariants: {
    id: string;
    content: string;
    checkType: 'keyword' | 'regex' | 'semantic';
    checkValue?: string;
    severity: 'warning' | 'error';
  }[];
}

Example Response:

{
  "invariants": [
    {
      "id": "inv_001",
      "content": "Never use console.log in production code",
      "checkType": "keyword",
      "checkValue": "console.log",
      "severity": "warning"
    },
    {
      "id": "inv_002",
      "content": "All API routes must have authentication middleware",
      "checkType": "semantic",
      "severity": "error"
    }
  ]
}

Best Practice: AI agents should call this before generating code to ensure they don’t violate any project rules.

`chorum_get_project_context`

Get high-level project metadata including tech stack, custom instructions, and confidence score.

Input Schema:

{
  projectId: string;
}

Output:

{
  name: string;
  description: string;
  techStack: string[];
  customInstructions: string;
  confidence: {
    score: number;           // 0-100
    interactionCount: number;
  };
  criticalFiles: string[];
}

Example Response:

{
  "name": "ChorumAI",
  "description": "Multi-provider LLM orchestration platform",
  "techStack": ["Next.js", "TypeScript", "Drizzle ORM", "PostgreSQL", "Supabase"],
  "customInstructions": "Prefer explicit TypeScript types. Use Zod for runtime validation.",
  "confidence": {
    "score": 78,
    "interactionCount": 247
  },
  "criticalFiles": ["src/lib/providers.ts", "src/lib/db/schema.ts"]
}

Write Tools

These tools modify your memory. They require human approval before changes take effect.

`chorum_propose_learning`

Propose a new pattern, decision, or learning item. The proposal is queued for user approval.

Input Schema:

{
  projectId: string;
  type: 'pattern' | 'antipattern' | 'decision' | 'invariant';
  content: string;              // The learning content
  context?: string;             // Why this was learned
  source: string;               // Agent identifier: "claude-code", "cursor", etc.
}

Output:

{
  proposalId: string;
  status: 'pending_approval';
  message: string;
}

Example Usage:

Agent notices you consistently use early returns. It proposes:

{
  "projectId": "proj_abc123",
  "type": "pattern",
  "content": "Use early returns to reduce nesting in handler functions",
  "context": "Observed in multiple code review sessions",
  "source": "claude-code"
}

Example Response:

{
  "proposalId": "prop_xyz789",
  "status": "pending_approval",
  "message": "Proposal queued. User will review in ChorumAI UI."
}

User Flow:

Agent calls chorum_propose_learning
Proposal appears in ChorumAI’s Pending Learnings section
User reviews, optionally edits, then approves or denies
If approved, the item is added to project memory

`chorum_log_interaction`

Log that an interaction occurred. Used for confidence scoring. Auto-approved (no human review required).

Input Schema:

{
  projectId: string;
  source: string;               // "claude-code", "cursor", "windsurf", etc.
  queryType: 'trivial' | 'moderate' | 'complex' | 'critical';
}

Output:

{
  success: boolean;
  newConfidenceScore: number;   // Updated project confidence
}

Example Response:

{
  "success": true,
  "newConfidenceScore": 79
}

Why This Matters: More interactions increase your project’s confidence score, which affects how much context is injected into future queries.

Learning Types

Type	Description	Example
`pattern`	Coding conventions, recurring approaches	”Use early returns to reduce nesting”
`antipattern`	Things to avoid	”Don’t use `any` type in TypeScript”
`decision`	Technical choices with rationale	”Chose PostgreSQL over SQLite for multi-user support”
`invariant`	Rules that must never be violated	”All API routes require auth middleware”

Rate Limits

Limit	Value
Requests per minute	100
Max tokens per query	8,000
Max pending proposals	50

Exceeding rate limits returns a 429 Too Many Requests error.

Error Responses

All errors follow this format:

{
  "error": {
    "code": "invalid_token",
    "message": "Token not found or revoked"
  }
}

Error Code	Meaning
`invalid_token`	Token is invalid, expired, or revoked
`project_not_found`	Project ID doesn’t exist or user lacks access
`rate_limited`	Too many requests
`invalid_input`	Missing required fields or invalid values

MCP Overview — What MCP is and why it matters
Setup Guide — Token generation and configuration
IDE Integration — Per-IDE setup instructions

Full Setup Guide IDE Integration

MCP Tools Reference

Why This Matters

Read Tools

chorum_list_projects

chorum_query_memory

chorum_get_invariants

chorum_get_project_context

Write Tools

chorum_propose_learning

chorum_log_interaction

Learning Types

Rate Limits

Error Responses

Related Documentation

`chorum_list_projects`

`chorum_query_memory`

`chorum_get_invariants`

`chorum_get_project_context`

`chorum_propose_learning`

`chorum_log_interaction`