M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
C

ContextStream MCP Server

Official
Persistent memory and semantic code search for AI coding assistants
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #10 of 438
Stars
★ 44 Popular
Transport
stdio (local process) · Streamable HTTP
Runtime
Node.js · Hosted (no local runtime)
Credentials
API key / credential required
License
MIT
Last commit
Tools
17
73FMRS · B

A feature-rich memory and code-search MCP server with a setup wizard covering major editors and a free tier. Its core value is cross-session memory, semantic search, and compaction protection, at the cost of cloud dependency and an API key.

Strongest · Documentation 18/20 Weakest · Reliability 10/20

Reliability
10/20
Security and permissions
15/20
Maintenance
15/20
Documentation
18/20
Setup experience
15/20
Why each score
Reliability 10/20
Based on static analysis, the MCP server shows basic structural integrity but lacks verifiable execution evidence. The server manifest indicates version 1.0.4 and clear tool definitions, but provides no evidence of CI workflows or test coverage for key paths. As this is a static analysis, we cannot verify the reliability of the startup and tool-handshake process.
Security and permissions 15/20
Credential handling is proper with the CONTEXTSTREAM_API_KEY environment variable marked as isSecret. Data boundaries are implemented through the workspace_id concept, and external network calls are configurable via API_URL. No obvious dangerous operation defaults were found, though detailed information on data encryption and access controls is lacking.
Maintenance 15/20
The project has 39 stars, 0 open issues, and is not archived. The MIT license is clear. Version history shows continuous updates, but lacks specific information on security response channels. The dependency update path is unclear as there's no package. file for review.
Documentation 18/20
Documentation is very detailed, including installation instructions, configuration examples, troubleshooting guides, and tool descriptions. Specific configuration instructions are provided for multiple editors. While comprehensive, the documentation lacks explicit disclosure of limitations and costs.
Setup experience 15/20
Provides setup commands supporting both interactive and zero-prompt modes. Configuration examples are provided for multiple editors, showing platform compatibility. However, as this is a static analysis, we cannot verify the smoothness of the actual installation process.

Static review · not runListed 2026-09-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesUses the network

Best for

  • Developers on long-running projects using Claude Code, Cursor, Windsurf, or similar tools
  • Teams needing cross-session memory and codebase understanding

Not for

  • Users unwilling to index code to a cloud service
  • One-off scripts that need no persistent context

Required permissions

  • Requires a ContextStream API key (secret env var)
  • Reads and indexes project code files
  • Optionally connects GitHub, Slack, and Notion integrations
  • Writes MCP configs, rules, and lifecycle hooks into editor settings

Risks and side effects

  • Indexed code is stored in ContextStream's cloud (vendor states isolation per workspace)
  • A leaked API key could expose workspace data
  • Breaking changes with @modelcontextprotocol/sdk 1.28.0+
  • Hosted HTTP OAuth may be blocked on corporate networks

Setup

Before you start

Runtime：Node.js · Hosted (no local runtime)

CONTEXTSTREAM_API_KEY requiredsecret ContextStream API key, required secret; get one by signing up at contextstream.io (free tier available).
Other optional settings (4)
CONTEXTSTREAM_API_URL optional ContextStream API base URL, optional, defaults to https://api.contextstream.io.
CONTEXTSTREAM_TOOLSET optional Controls which toolset is registered; example configs set it to complete.
CONTEXTSTREAM_HOOK_ENABLED optional Set to false to disable editor lifecycle hooks.
CONTEXTSTREAM_VSCODE_MCP_MODE optional Set to local to force VS Code/Copilot to use a local runtime instead of the hosted remote server.

Run npx --prefer-online -y @contextstream/mcp-server@latest setup to auto-configure editors, or manually add the npx launch command to your MCP config with CONTEXTSTREAM_API_KEY (required) and CONTEXTSTREAM_API_URL (optional, default https://api.contextstream.io). A hosted remote endpoint at https://mcp.contextstream.io/mcp (streamable-http, with OAuth) is also available.

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

Shown for Claude Code. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "contextstream": {
      "command": "npx",
      "args": [
        "--prefer-online",
        "-y",
        "@contextstream/mcp-server@latest"
      ],
      "env": {
        "CONTEXTSTREAM_API_URL": "https://api.contextstream.io",
        "CONTEXTSTREAM_API_KEY": "your_key"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add contextstream -e CONTEXTSTREAM_API_URL=https://api.contextstream.io -e CONTEXTSTREAM_API_KEY=your_key -- npx --prefer-online -y @contextstream/mcp-server@latest

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Run `npx --prefer-online -y @contextstream/mcp-server@latest setup`, restart your editor, and confirm the contextstream tools (e.g. search, context, memory) appear in your client's tool list, or call help(action="version") and check it returns version info.

Troubleshooting

  1. Run contextstream-mcp doctor to check auth, index health, rules, and hooks
  2. Verify CONTEXTSTREAM_API_URL and CONTEXTSTREAM_API_KEY are set
  3. Remove stale version pins like @contextstream/[email protected]
  4. Restart the editor after config changes
  5. Use stdio with an API key if OAuth is blocked
  6. Check SDK resolution for Zod schema errors

Things to try

Once connected, you can ask your AI assistant things like:

  • Where do we handle authentication?
  • What did we decide about the id format last week?
  • What depends on UserService?
  • What breaks if I change this function?

Tools 17

init writes
Get workspace state and prepare context for every message
context read-only
Deliver the right workspace context on every message
search read-only
Code search with semantic, hybrid, keyword, pattern, exhaustive, and refactor modes
memory writes
Manage events, decisions, docs, runbooks, tasks, todos, diagrams, and transcripts
session writes
Capture decisions and lessons, recall past sessions and plans
qa read-only
Grounded Q&A over your workspace knowledge base with citations
graph read-only
Dependency mapping, impact analysis, circular-dependency and dead-code detection
capsule writes
Package portable, shareable context snapshots for agent handoffs
Show 9 more tools
entity writes
Manage tickets, incidents, releases, sprints, OKRs, and risks
project writes
Project indexing, scope, and de-indexing management
workspace writes
Workspace management
skill writes
Reusable instruction and action bundles, portable across tools
media writes
Index and search images, video, audio, and documents
vcs read-only
Repository links management
reminder writes
Reminders
integration writes
Third-party integrations
help read-only
Diagnostics and version information

Use cases

Remember decisions, plans, and lessons across sessions
Semantic search over large codebases
Pre-compaction capture of critical state before context loss
Team knowledge integration with cited Q&A
Context handoff between agents

Supported clients

Claude Code
Cursor
Windsurf
Cline
Roo Code
VS Code + GitHub CopilotPartial support
Kilo CodePartial support
CodexPartial support
OpenCodePartial support
AiderPartial support
AntigravityPartial support
Claude DesktopPartial support

Listed from the project's documentation, not tested by this site.

Overview

ContextStream MCP Server gives MCP-compatible AI coding assistants (Claude Code, Cursor, VS Code Copilot, Windsurf, and more) persistent memory, semantic code search, and cross-session learning. It indexes your codebase, records decisions and lessons, maps the dependency graph, and pulls in team knowledge from GitHub, Slack, and Notion. Per vendor benchmarks it scores 90.0% on the full 500-instance LongMemEval-S suite with the official GPT-4o judge.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 83eee4ee54b5 Data synced 2026-10-19 Read the FMRS scoring method