Skip to main content
Glama
contextstream

ContextStream MCP Server

by contextstream
OverviewSchemaRelated ServersScoreDiscussions
TypeScript
Remote

ContextStream MCP Server

Persistent memory, semantic search, and code intelligence for any MCP-compatible AI tool (Cursor, Claude Code, Windsurf, VS Code, Claude Desktop, Codex CLI, etc.).

ContextStream is a shared "brain" for AI-assisted development. It stores decisions, preferences, lessons, tasks, and project context — and lets your AI tools search and analyze your codebase with consistent context across sessions.

v0.4.x: Consolidated domain tools (~11 tools) for ~75% lower tool-registry token overhead vs previous versions.

Just Ask

You don't need to memorize tool names. Describe what you want and your AI will call the right ContextStream tools:

You say…

ContextStream does…

"session summary"

Summarizes current workspace/project context

"what did we decide about auth?"

Recalls decisions related to authentication

"remember we're using PostgreSQL"

Captures that fact for future sessions

"search for payment code"

Searches your codebase semantically/hybrid/keyword

"what depends on UserService?"

Analyzes dependency graph & impact

Tip: For best results, add the recommended editor rules so your AI reliably calls session_init / context_smart when appropriate: https://contextstream.io/docs/quickstart

ContextStream in action

Choose Your Mode (Token Footprint)

MCP clients often inject the tool catalog into the model context. v0.4.x is designed to keep that overhead small.

Mode

What it exposes

Best for

Enable

Consolidated (default)

~11 domain tools with action / mode dispatch

Most users (recommended)

CONTEXTSTREAM_CONSOLIDATED=true

Router (extreme minimization)

~2 meta-tools (contextstream, contextstream_help)

Tight context budgets / many MCP servers

CONTEXTSTREAM_PROGRESSIVE_MODE=true

Legacy (granular tools)

Older light/standard/complete toolsets

Back-compat / old prompts

CONTEXTSTREAM_CONSOLIDATED=false

Note: The env var name for Router mode is CONTEXTSTREAM_PROGRESSIVE_MODE (historical naming). It enables the ~2-tool "router" surface.

Features

  • Consolidated domain tools (v0.4.x): short tool list with action/mode dispatch

  • Session-aware context loading (session_init, context_smart)

  • Memory capture + recall (decisions, preferences, lessons, tasks, bugs)

  • Code search (semantic, hybrid, keyword, pattern)

  • Knowledge graph + code analysis (dependencies, impact, call paths, circular deps, unused code)

  • Graph ingestion for full graph builds (graph(action="ingest"))

  • Local repo ingestion for indexing (project(action="ingest_local"))

  • Auto-context: on first tool call in a new session, the server can auto-initialize context

Graph Tiers

Tier

Capabilities

Pro (Graph-Lite)

Module-level import graph, dependencies, and 1-hop impact

Elite/Team (Full Graph)

Module + call + dataflow + type layers, plus full graph ingestion

Requirements

  • Node.js 18+

  • A ContextStream account

  • Auth via API key or JWT

Default API URL: https://api.contextstream.io

Quickstart (2 minutes)

1) Run the Setup Wizard (recommended)

The wizard:

  • Authenticates (browser/device login by default)

  • Creates/stores an API key

  • Installs recommended editor rules (optional)

  • Writes MCP config files for supported tools

npx -y @contextstream/mcp-server setup

Useful flags:

Flag

Description

--dry-run

Preview without writing files

Notes:

  • The wizard stores credentials at ~/.contextstream/credentials.json for convenience. Delete it to force a fresh login.

  • Codex CLI MCP config is global-only (~/.codex/config.toml), so the wizard writes Codex config globally when selected.

  • Some tools still require UI/CLI setup (the wizard will tell you when it can't write a config).

2) Run the MCP Server

Recommended (works well with MCP configs):

npx -y @contextstream/mcp-server

Or install globally:

npm install -g @contextstream/mcp-server contextstream-mcp

3) Keeping Updated

If you use Restart your AI tool/editor and run ContextStream again (or pin the version: npx -y @contextstream/mcp-server@0.4.3)

If you installed globally:

npm update -g @contextstream/mcp-server

After updating, restart your AI tool/editor so it reloads the tool catalog.

Configure Your MCP Client (Manual)

If you ran the setup wizard, you can usually skip this.

Cursor / Windsurf / Claude Desktop (JSON)

These clients use an mcpServers JSON config:

Client

Config path

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json (project)

Windsurf

~/.codeium/windsurf/mcp_config.json

Claude Desktop (macOS)

~/Library/Application Support/Claude/claude_desktop_config.json

Claude Desktop (Windows)

%APPDATA%\Claude\claude_desktop_config.json

Consolidated (default):

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

Router mode (~2 meta-tools):

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

VS Code (.vscode/mcp.json)

VS Code uses a top-level servers map:

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

Strong recommendation: Use inputs so you don't commit secrets:

{ "servers": { "contextstream": { "type": "stdio", "command": "npx", "args": ["-y", "@contextstream/mcp-server"], "env": { "CONTEXTSTREAM_API_URL": "https://api.contextstream.io", "CONTEXTSTREAM_API_KEY": "${input:contextstreamApiKey}" } } }, "inputs": [ { "id": "contextstreamApiKey", "type": "promptString", "description": "ContextStream API Key", "password": true } ] }

Claude Code (CLI)

User scope (all projects):

claude mcp add --transport stdio contextstream --scope user \ --env CONTEXTSTREAM_API_URL=https://api.contextstream.io \ --env CONTEXTSTREAM_API_KEY=YOUR_KEY \ -- npx -y @contextstream/mcp-server

Router mode:

claude mcp add --transport stdio contextstream --scope user \ --env CONTEXTSTREAM_API_URL=https://api.contextstream.io \ --env CONTEXTSTREAM_API_KEY=YOUR_KEY \ --env CONTEXTSTREAM_PROGRESSIVE_MODE=true \ -- npx -y @contextstream/mcp-server

Windows caveat (native Windows, not WSL): if npx isn't found, use cmd /c npx -y @contextstream/mcp-server after --.

Codex CLI (~/.codex/config.toml)

[mcp_servers.contextstream] command = "npx" args = ["-y", "@contextstream/mcp-server"] [mcp_servers.contextstream.env] CONTEXTSTREAM_API_URL = "https://api.contextstream.io" CONTEXTSTREAM_API_KEY = "YOUR_API_KEY"

Tool Overview (v0.4.x Consolidated)

In consolidated mode, you call domain tools with action / mode:

Core

Tool

Description

session_init

Initialize workspace/project context

context_smart

Retrieve the best bounded context for the current message

Domain Tools

Tool

Description

search

mode=semantic|hybrid|keyword|pattern

session

action=capture|recall|remember|get_lessons|capture_lesson|...

memory

Events + nodes CRUD, decisions, lessons, etc.

graph

Dependencies, impact, call_path, ingest, etc.

project

Indexing, ingest_local, stats, files, etc.

workspace

List, get, associate, bootstrap

reminder

List, create, snooze, complete, dismiss

integration

provider=slack|github, search, activity, etc.

help

Tools, auth, version, editor_rules

Examples

search(mode="semantic", query="auth middleware") memory(action="create_node", node_type="decision", title="Auth strategy", content="...") graph(action="impact", target="UserService")

Full tool catalog: https://contextstream.io/docs/mcp/tools

Authentication

Set one of:

Variable

Use case

CONTEXTSTREAM_API_KEY

Recommended for local/dev

CONTEXTSTREAM_JWT

Useful for hosted/user-session flows

Environment Variables

Required

Variable

Description

CONTEXTSTREAM_API_URL

Base API URL (default https://api.contextstream.io)

CONTEXTSTREAM_API_KEY

API key (unless using JWT)

CONTEXTSTREAM_JWT

JWT (unless using API key)

Token + Tool Surface Controls

Variable

Description

CONTEXTSTREAM_CONSOLIDATED

true (default in v0.4.x) uses consolidated domain tools

CONTEXTSTREAM_PROGRESSIVE_MODE

Enables Router mode (~2 meta-tools)

CONTEXTSTREAM_CONTEXT_PACK

Enable Context Pack for context_smart (code + graph + distill). Defaults to true

CONTEXTSTREAM_TOOLSET

Legacy granular tool bundle: light / standard / complete (only when consolidated is off)

CONTEXTSTREAM_TOOL_ALLOWLIST

Comma-separated tool names to expose (legacy granular mode)

CONTEXTSTREAM_SCHEMA_MODE

Reduce schema verbosity; e.g., compact

CONTEXTSTREAM_OUTPUT_FORMAT

Output formatting; e.g., compact / pretty

Optional Defaults

Variable

Description

CONTEXTSTREAM_WORKSPACE_ID

Default workspace fallback

CONTEXTSTREAM_PROJECT_ID

Default project ID fallback

CONTEXTSTREAM_USER_AGENT

Custom user agent string

CONTEXTSTREAM_PRO_TOOLS

Comma-separated tool names treated as PRO

CONTEXTSTREAM_UPGRADE_URL

Upgrade link for Free users calling PRO tools

Migration Notes (pre-0.4.x → 0.4.x)

Most workflows just work, but tool names change in consolidated mode.

Before (granular)

After (consolidated)

search_semantic(query="auth")

search(mode="semantic", query="auth")

session_capture(...)

session(action="capture", ...)

graph_dependencies(...)

graph(action="dependencies", ...)

If you rely on granular tool names, you can temporarily set:

CONTEXTSTREAM_CONSOLIDATED=false

Troubleshooting

Issue

Solution

Tools not appearing

Restart the client after editing MCP config; confirm Node 18+ is available in the client runtime

Unauthorized / 401

Verify CONTEXTSTREAM_API_URL + CONTEXTSTREAM_API_KEY (or JWT)

Wrong workspace/project

Run session_init and/or associate your folder with the correct workspace

Client warns about tool context size

Use Router mode (CONTEXTSTREAM_PROGRESSIVE_MODE=true), or keep consolidated mode and reduce schema/output verbosity

Development

git clone https://github.com/contextstream/mcp-server.git cd mcp-server npm install npm run dev npm run typecheck npm run build

Links

Resource

URL

Website

https://contextstream.io

Docs

https://contextstream.io/docs/mcp

Tool Catalog

https://contextstream.io/docs/mcp/tools

Pricing

https://contextstream.io/pricing

npm

https://www.npmjs.com/package/@contextstream/mcp-server

GitHub

https://github.com/contextstream/mcp-server

License

MIT

This server cannot be installed

-
security - not tested
A
license - permissive license
-
quality - not tested

How are these scores calculated?

Resources

New MCP Servers

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/contextstream/mcp-server'

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