Skip to content
/

context-space

context-space · context-space/context-space · ★ 810 · last commit 2025-10-22

Primitive shape 1 total
MCP tools 1
00

Summary

context-space — Summary

Context Space is a Go-based open-source backend infrastructure project that provides a unified MCP server aggregating OAuth-secured integrations with 14+ external services (GitHub, Slack, Airtable, HubSpot, Notion, Spotify, Stripe, etc.) for AI agents. It positions itself as "the first context engineering infrastructure" — turning the theory of context engineering into production-ready tooling. The system manages credentials via HashiCorp Vault, provides one-click OAuth flows (no JSON config editing), exposes a unified RESTful API at api.context.space, and offers Cursor IDE deep-link installation. Current phase delivers secure credential management and tool aggregation; future phases plan native MCP support, context memory, semantic retrieval, and context optimization.

Compared to seeds: no seed is a hosted MCP aggregator infrastructure project. The closest is ccmemory (MCP tool access for AI agents) but context-space is infrastructure (server backend + credential management) rather than a memory system. Unlike ccmemory which focuses on session memory, context-space focuses on real-time external service integration with enterprise security (HashiCorp Vault, AGPL license transitioning to Apache 2.0).

01

Overview

context-space — Origin & Philosophy

Origin

Created by the context-space organization, 14 contributors, AGPL-3.0 license (transitioning to Apache 2.0). Go-primary codebase. 810 stars. Last commit 2025-10-22.

Mission

"Context Space: The First Context Engineering Infrastructure to 10× Your Productivity"

"Today's AI agents excel at reasoning but terrible at acting in the real world. They're cut off from live data and tools, trapped behind scattered APIs, inconsistent sources, and complex authentication."

Core Value Proposition

Context Space solves three production problems with MCP in enterprise contexts:

  1. Manual, Insecure Credential Handling → One-Click OAuth + HashiCorp Vault
  2. Inconsistent and Complex APIs → Single unified RESTful API
  3. Complex Deployment and Scattered MCP Servers → Unified Context Plane with Tool Aggregation

Philosophy

Context Space frames itself as the bridge from "context engineering theory" to practice:

"Context engineering is the foundation for building reliable AI agents. It goes beyond prompt engineering by managing not only what users say to the model, but also the broader context that shapes its behavior, such as tools, memory, and data."

"MCP defines a standard path for agents to securely access real-world services. Context Space brings that vision to life by turning MCP into production-ready infrastructure."

License Strategy

AGPL-3.0 now (startup protection phase) → Apache 2.0 future (community growth phase). CLA required for contributors to enable this transition.

Roadmap

Phase 1 (Available Now)

Production-ready: OAuth + Vault, unified API, tool aggregation, 14+ integrations

Phase 2 (6-12 months)

Native MCP Support, Context Memory, Smart Aggregation, Semantic Retrieval, Context Optimization

Phase 3 (12+ months)

Context Synthesis, Predictive Loading, AI Context Reasoning

02

Architecture

context-space — Architecture

Distribution

  • Type: standalone-repo (Go backend server + React web frontend)
  • License: AGPL-3.0 (transitioning to Apache 2.0)
  • Backend language: Go 1.24
  • Storage: PostgreSQL + HashiCorp Vault
  • Deploy: Docker + systemd

Install Methods

# Via Claude Code CLI
claude mcp add "context-space" https://api.context.space/api/mcp --header "Authorization: Bearer YOUR_API_KEY"

# Via Cursor IDE deep link (one-click)
# cursor://context-space/<integration>  — no JSON editing

# Self-host (Docker)
docker compose -f backend/docker/docker-compose.yml up

# Get API key at: https://app.context.space/settings

Required Runtime (Self-hosted)

  • Go 1.24
  • Docker
  • PostgreSQL
  • HashiCorp Vault

Directory Structure

context-space/
├── backend/
│   ├── cmd/
│   │   ├── server/           # HTTP server entry point
│   │   ├── mcp-tool/         # MCP server binary
│   │   ├── load_providers/   # Integration loader
│   │   └── credentialmanagement/
│   ├── internal/
│   │   ├── credentialmanagement/  # Vault-backed credential storage
│   │   ├── identityaccess/        # Auth/authz
│   │   ├── integration/           # OAuth integration handlers
│   │   ├── provideradapter/       # Service adapters (GitHub, Slack, etc.)
│   │   ├── providercore/          # Provider abstraction
│   │   └── translation/           # API format translation
│   ├── hashivault/           # HashiCorp Vault integration
│   ├── migrations/           # Database migrations
│   └── docker/
├── web/                      # Frontend (unknown stack)
└── docs/

Target AI Tools

  • Claude Code (MCP via claude mcp add)
  • Cursor IDE (deep links + MCP)
  • Any MCP client (hosted endpoint api.context.space/api/mcp)
  • Claude Desktop, Windsurf (via mcp.json config)

Hosted Service

Available at:

  • MCP endpoint: https://api.context.space/api/mcp
  • API: https://api.context.space/v1/
  • Web UI: https://app.context.space/
  • Integrations browser: https://context.space/integrations
03

Components

context-space — Components

MCP Server (1)

Single unified MCP server exposing all integrated services as MCP tools.

Endpoint Description
https://api.context.space/api/mcp Hosted MCP endpoint (Streamable HTTP)

All 14+ integrations are surfaced as individual MCP tool calls through this single endpoint.

Production-Ready Integrations (14+)

Service Category Auth Status
GitHub Development OAuth Ready
Slack Communication OAuth Ready
Airtable Data Management OAuth Ready
HubSpot CRM OAuth Ready
Notion Knowledge OAuth Ready
Spotify Personal OAuth Ready
Stripe Financial API Key Ready
7+ more Various Various Ready

REST API

https://api.context.space/v1/ provides:

  • GET /users/me — Authentication check
  • POST /credentials/auth/oauth/{service}/auth-url — Get OAuth URL
  • POST /invocations/{service}/{operation} — Execute tool operations
  • Full Swagger docs at http://api.context.space/v1/docs

Backend Components (Go)

Component Purpose
credentialmanagement HashiCorp Vault-backed credential storage
identityaccess Authentication and authorization
integration OAuth flow handlers
provideradapter Service-specific API adapters
providercore Abstract provider interface
translation API format normalization
hashivault Vault client library

Slash Commands / Skills

None — this is backend infrastructure, not an agent framework.

Hooks

None documented.

Scripts

Makefile targets for lint, tests, Docker builds.

05

Prompts

context-space — Prompt Files (Verbatim Excerpts)

Context Space is a backend infrastructure project with no agent prompt files, skills, or instruction documents beyond a brief API usage guide. The "prompts" it works with are natural language instructions to AI agents that trigger MCP tool calls.

Excerpt 1: README API Usage Examples

Source: README.md

Technique: Operational examples showing how agents interact with Context Space via natural language → MCP tool calls. Not a prompt file, but demonstrates the interaction pattern.

# Authentication
curl -H "Authorization: Bearer <jwt-token>" \
     https://api.context.space/v1/users/me

# Create OAuth Authorization URL
curl -H "Authorization: Bearer <jwt-token>" \
     -X POST \
     https://api.context.space/v1/credentials/auth/oauth/github/auth-url

# Execute Operations (e.g., list GitHub repos)
curl -H "Authorization: Bearer <jwt-token>" \
     -X POST \
     https://api.context.space/v1/invocations/github/list_repositories

Excerpt 2: Conceptual Framing

Source: README.md

Technique: Positioning statement that frames Context Space as "context engineering infrastructure" — useful for understanding how it fits into agent architectures.

Context engineering is the foundation for building reliable AI agents. 
It goes beyond prompt engineering by managing not only what users say 
to the model, but also the broader context that shapes its behavior, 
such as tools, memory, and data.

MCP defines a standard path for agents to securely access real-world services. 
Context Space brings that vision to life by turning MCP into production-ready 
infrastructure.

Note

This framework has no skill files, command files, or prompt templates. Its contribution is server-side infrastructure, not agent instruction design.

09

Uniqueness

context-space — Uniqueness & Positioning

differs_from_seeds

No seed framework is a hosted MCP aggregator infrastructure project. The closest is ccmemory (MCP-anchored tool access for Claude Code) but ccmemory is a self-hosted memory system while context-space is production infrastructure for real-time external service access. Context-space uses HashiCorp Vault for credential management — an enterprise security primitive not present in any seed. Unlike Nex (which integrates similar external services via cloud KB), context-space exposes them as MCP tools for real-time invocation rather than knowledge graph queries. The Go backend + PostgreSQL + Vault stack is the most "production-grade" technical architecture in this batch.

Positioning

Context-space is uniquely positioned as infrastructure-as-service for MCP — it handles the hardest part of MCP adoption (OAuth credential management, service authentication, API normalization) so AI agents can access real-world services without friction. The AGPL→Apache license transition signals a startup that wants to build community before commercializing.

Observable Failure Modes

  • Cloud dependency: All credentials managed by Context Space's cloud infrastructure. Self-hosting requires Go, Docker, and HashiCorp Vault.
  • 810 stars, last commit 2025-10: Lower engagement than its positioning implies. The roadmap's Phase 2 (memory, semantic retrieval) is ambitious; delivery is unconfirmed.
  • No memory in current release: Despite calling itself "context engineering infrastructure," current Phase 1 has no memory, no retrieval, no context optimization — just OAuth-secured tool aggregation.
  • AGPL license risk: Enterprise users may be blocked by AGPL until Apache 2.0 transition occurs.

Cross-References

  • Conceptually related to: Nex (also aggregates external services for AI agents)
  • More infrastructure-oriented than: contextmcp (similar positioning, no public code)
04

Workflow

context-space — Workflow

Phase 1: Connect (Claude Code)

# Get API key at https://app.context.space/settings
claude mcp add "context-space" https://api.context.space/api/mcp --header "Authorization: Bearer YOUR_API_KEY"

Artifact: MCP server registered in Claude Code

Phase 2: Authenticate Services

Use the web UI at https://app.context.space/integrations or via API:

curl -H "Authorization: Bearer <jwt-token>" \
     -X POST \
     https://api.context.space/v1/credentials/auth/oauth/github/auth-url

OAuth flow handled by Context Space — credentials stored in HashiCorp Vault.

Artifact: Service credentials securely stored, no manual JSON editing

Phase 3: Agent Use

AI agent calls MCP tools to interact with connected services:

"List my GitHub repositories"
"Post to Slack channel #engineering"
"Create a Notion page"
"Query HubSpot for contacts"

All operations go through the unified https://api.context.space/api/mcp endpoint.

Artifact: Real-world actions executed on connected services

Phase 4: Direct API Access (for custom integrations)

curl -H "Authorization: Bearer <jwt-token>" \
     -X POST \
     https://api.context.space/v1/invocations/github/list_repositories

Approval Gates

None — agent executes operations directly through MCP tools.

Artifacts Summary

Artifact Description
OAuth credentials Stored in HashiCorp Vault
API key At app.context.space/settings
MCP tool responses Real-time data from connected services
06

Memory Context

context-space — Memory & Context

Current State

Phase 1 has no memory system — it is a real-time tool execution layer, not a memory store.

Planned (Phase 2: 6-12 months)

From the roadmap:

  • Context Memory — persistent memory for AI agents
  • Semantic Retrieval — search over stored context
  • Smart Aggregation — intelligent context selection
  • Context Optimization — trimming context window payload

Phase 3 Plans (12+ months)

  • Context Synthesis — combining multiple sources intelligently
  • Predictive Loading — preemptive context retrieval
  • AI Context Reasoning — context decisions made by AI

Current Context Contribution

Context Space contributes live external data to agent context — real-time GitHub repos, Slack messages, HubSpot CRM records. This is "context" in the sense of grounding agents in current real-world state, not episodic memory of past sessions.

State Files

No local state files. All state:

  • Credentials: HashiCorp Vault (cloud)
  • Integration config: api.context.space (cloud)
  • API key: User manages locally
07

Orchestration

context-space — Orchestration

Multi-Agent

No — Context Space is called by agents, not an orchestrator.

Orchestration Pattern

none

Execution Mode

event-driven — responds to MCP tool calls from AI agents on demand.

Isolation Mechanism

Server-side isolation via HashiCorp Vault (credential isolation between users/projects).

Multi-Model

No — model-agnostic infrastructure.

Crash Recovery

Unknown — server-side resilience handled by Docker + systemd deployment.

Cross-Tool Portability

High — any MCP client can use the hosted endpoint. Cursor has deep-link support.

08

Ui Cli Surface

context-space — UI & CLI Surface

Dedicated CLI Binary

None — integration is via claude mcp add command or MCP JSON config.

Local Web Dashboard

No local dashboard. Hosted web UI at https://app.context.space/ and https://context.space/integrations.

Web UI Features

  • OAuth connection management (connect GitHub, Slack, etc.)
  • Integration browsing
  • API key management
  • One-click OAuth flows (no JSON editing)

Cursor IDE Integration

Deep-link support: cursor:// deep links for one-click installation — no JSON file editing required.

API Documentation

Swagger/OpenAPI docs at http://api.context.space/v1/docs

MCP Protocol Support

  • Hosted Streamable HTTP: https://api.context.space/api/mcp
  • Local self-hosted: backend/cmd/mcp-tool/

Observability

Server-side only (logs, monitoring not documented in public repo).

Related frameworks

same archetype · same primary tool · same memory type

claude-mem (thedotmack) ★ 78k

Background worker service captures every tool call as an observation, AI-compresses sessions, and auto-injects relevant past…

pi (badlogic/earendil) ★ 55k

A minimal, hackable, multi-provider terminal coding agent that adapts to your workflows via npm-installable TypeScript Extensions…

Agent Skills (Addy Osmani) ★ 46k

Encodes senior-engineer software development lifecycle as 23 auto-routed skills and 7 slash commands for any AI coding agent.

wshobson/agents Plugin Marketplace ★ 36k

Single Markdown source for 83 domain-specialized plugins that auto-generates idiomatic artifacts for five AI coding harnesses.

TabbyML/Tabby ★ 34k

Self-hosted AI coding assistant server (alternative to GitHub Copilot) with admin dashboard, RAG-based completions, and multi-IDE…

Compound Engineering ★ 17k

Make each unit of engineering work compound into easier future work via brainstorm→plan→execute→review→learn cycles.