Package Exports

@contextstream/mcp-server
@contextstream/mcp-server/dist/index.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@contextstream/mcp-server) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

ContextStream MCP Server

A powerful Model Context Protocol (MCP) server that provides persistent memory, code context, knowledge graphs, and AI tools. Connect your AI assistant to your codebase's semantic search, memory system, and dependency analysis.

Website: contextstream.io | Docs: contextstream.io/docs/mcp

Features

🔍 Code Search & Context

Semantic search — Find code by meaning, not just keywords
Hybrid search — Combines semantic + keyword for best results
Pattern search — Regex-based code search
AI context building — Automatically gather relevant code, docs, and memory

🧠 Memory & Knowledge

Memory events — Store decisions, insights, and context
Knowledge nodes — Build a graph of connected concepts
Decision tracking — Never lose architectural decisions
Memory search — Find relevant past context

📊 Code Intelligence

Dependency analysis — Understand what depends on what
Call path tracing — Trace execution flows
Impact analysis — Understand change impact
Circular dependency detection — Find problematic cycles
Unused code detection — Identify dead code

🤖 AI Integration

Context building — Build LLM-ready context from your codebase
Enhanced context — Deep analysis with memory integration
Plan generation — AI-powered development planning
Task generation — Break down work into actionable tasks

Installation

Add ContextStream to your AI tool's MCP configuration (works with Cursor, VS Code, Claude Code, Windsurf, and any MCP-compatible client):

{
  "mcpServers": {
    "contextstream": {
      "command": "npx",
      "args": ["-y", "@contextstream/mcp-server"],
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_api_key"
      }
    }
  }
}

Get your API key at contextstream.io

Codex (CLI) configuration

For the Codex CLI, add the MCP server to your ~/.codex/config.toml:

[mcpServers.contextstream]
command = "npx"
args = ["-y", "@contextstream/mcp-server"]

[mcpServers.contextstream.env]
CONTEXTSTREAM_API_URL = "https://api.contextstream.io"
CONTEXTSTREAM_API_KEY = "your_api_key"

After saving the file, restart Codex. The ContextStream tools and auto-context will then be available in any trusted project that has a .contextstream/config.json (or a matching parent mapping in ~/.contextstream-mappings.json).

✨ Auto-Context (v0.3.0+)

Context loads automatically on the first tool call of any conversation. No need to manually call session_init — your AI assistant gets workspace info, recent memory, and decisions immediately.

═══════════════════════════════════════════
🧠 AUTO-CONTEXT LOADED (ContextStream)
═══════════════════════════════════════════
📁 Workspace: my-project
📂 Project: backend-api

📋 Recent Decisions:
   • Use PostgreSQL for persistence
   • JWT for authentication

🧠 Recent Context:
   • [decision] API design patterns
   • [preference] TypeScript strict mode
═══════════════════════════════════════════

Works with all MCP clients — no client-side changes required.

Environment Variables

Variable	Required	Description
`CONTEXTSTREAM_API_URL`	Yes	API base URL (`https://api.contextstream.io`)
`CONTEXTSTREAM_API_KEY`	Yes	Your API key from contextstream.io

Available Tools

Authentication

Tool	Description
`auth_me`	Get current user profile

Workspaces

Tool	Description
`workspaces_list`	List accessible workspaces
`workspaces_get`	Get workspace details
`workspaces_create`	Create a new workspace
`workspaces_overview`	Get workspace summary
`workspaces_analytics`	Get usage analytics
`workspaces_content`	List workspace content

Projects

Tool	Description
`projects_list`	List projects (by workspace)
`projects_get`	Get project details
`projects_create`	Create a new project
`projects_overview`	Get project summary
`projects_statistics`	Get code statistics
`projects_files`	List indexed files
`projects_index`	Trigger re-indexing
`projects_index_status`	Check indexing status

Search

Tool	Description
`search_semantic`	Semantic vector search
`search_hybrid`	Combined semantic + keyword
`search_keyword`	Traditional keyword search
`search_pattern`	Regex pattern search
`search_suggestions`	Get search suggestions

Memory

Tool	Description
`memory_create_event`	Store a memory event
`memory_get_event`	Retrieve an event
`memory_update_event`	Update an event
`memory_delete_event`	Remove an event
`memory_list_events`	List events in workspace
`memory_bulk_ingest`	Bulk import events
`memory_distill_event`	Extract key insights
`memory_create_node`	Create knowledge node
`memory_get_node`	Get knowledge node
`memory_update_node`	Update knowledge node
`memory_delete_node`	Remove knowledge node
`memory_supersede_node`	Replace with new version
`memory_list_nodes`	List knowledge nodes
`memory_search`	Search memory
`memory_decisions`	Get decision summaries
`memory_timeline`	Get chronological timeline
`memory_summary`	Get condensed summary

Graph Analysis

Tool	Description
`graph_related`	Find related knowledge nodes
`graph_path`	Find path between nodes
`graph_decisions`	Decision history in graph
`graph_dependencies`	Query code dependencies
`graph_call_path`	Trace call paths
`graph_impact`	Analyze change impact
`graph_circular_dependencies`	Find circular deps
`graph_unused_code`	Find dead code
`graph_contradictions`	Find conflicting info

AI

Tool	Description
`ai_context`	Build LLM context
`ai_enhanced_context`	Deep context with memory
`ai_embeddings`	Generate embeddings
`ai_plan`	Generate dev plan
`ai_tasks`	Generate tasks

Session & Auto-Context (CORE-like)

These tools enable automatic context initialization for AI assistants:

Tool	Description
`session_init`	Initialize conversation - First tool to call, returns workspace/project info, recent memory, decisions
`session_get_user_context`	Get user preferences and coding style from memory
`session_capture`	Store context (decisions, insights, preferences) to memory
`session_smart_search`	Search with automatic context enrichment (memory + code + decisions)
`session_remember`	Quick natural language memory storage ("Remember I prefer TypeScript strict mode")
`session_recall`	Quick natural language recall ("What were the auth decisions?")

Usage in AI Tools:

# In Cursor/Claude Code/Windsurf - start with:
"Initialize session for this workspace"

# Then use natural commands:
"Remember we decided to use PostgreSQL for the database"
"Recall the authentication decisions we made"
"What are my coding preferences?"

Prompts

Pre-built prompts for common workflows:

Prompt	Description
`explore-codebase`	Get a codebase overview
`capture-decision`	Document an ADR
`review-context`	Build code review context
`investigate-bug`	Debug investigation helper
`explore-knowledge`	Navigate knowledge graph
`onboard-to-project`	Generate onboarding guide
`analyze-refactoring`	Find refactoring opportunities
`build-context`	Build comprehensive LLM context

Resources

Resource	Description
`contextstream:openapi`	OpenAPI specification
`contextstream:workspaces`	Workspace listing
`contextstream:projects/{id}`	Projects for workspace

Error Handling

The client includes:

Automatic retries for transient failures (429, 500, 502, 503, 504)
Exponential backoff with configurable delays
Detailed error codes (NETWORK_ERROR, UNAUTHORIZED, RATE_LIMITED, etc.)
Request timeouts (180s default)

Security

No secrets are bundled — provide your own API key/JWT
No internal endpoints hardcoded — API URL is required
Minimal logging — credentials are not echoed
Public REST API only — no proprietary logic included

License

MIT