Claude Context MCP Server

The Claude Context MCP Server is a Cloudflare Worker that exposes Arda’s skill library, agent profiles, and documentation site content over the Model Context Protocol (MCP). It gives claude.ai cloud sessions access to the same skills, personas, and project knowledge that Claude Code CLI sessions get from local files.

Why

Claude Code sessions have direct filesystem access to the workspace/ repository — skills, agent profiles, and documentation are loaded on demand from local files. Cloud sessions on claude.ai have no such access. The MCP server bridges this gap by serving the same content over HTTP.

Architecture

claude.ai / Claude Code
        │
        │ MCP HTTP+SSE
        │ Authorization: Bearer <Base64(GitHub PAT)>
        ▼
┌──────────────────────────┐
│  Cloudflare Worker        │
│  claude-context-mcp       │
│                           │
│  list_* → bundled indexes │ ← built at deploy time (zero runtime API calls)
│  get_*  → GitHub API      │ ← live fetch with 10-min cache
└──────────────────────────┘
        │
        │ Authorization: Bearer <decoded PAT>
        ▼
┌──────────────────────────┐
│  GitHub Contents API      │
│  Arda-cards/agentic-      │
│    workspace              │
│  Arda-cards/documentation │
└──────────────────────────┘

Key design points:

list_* tools serve pre-built indexes bundled into the Worker at deploy time — zero runtime GitHub API calls.
get_* tools fetch live from GitHub with a 10-minute Cloudflare Cache API TTL. Pass skipCache: true to force a fresh fetch.
The Worker holds no secrets. The caller’s GitHub PAT travels in the request header, is decoded from URL-safe Base64, and forwarded to GitHub.
Durable Objects provide MCP session state.

Available Tools

Tool	Parameters	Description
`list_skills`	none	Lists all skills with names, descriptions, argument hints, and companion file manifests
`get_skill`	`name`, `skipCache?`	Fetches the full `SKILL.md` content for a named skill
`get_skill_file`	`skill`, `file`, `skipCache?`	Fetches an arbitrary file from a skill directory (returned Base64-encoded)
`list_agents`	none	Lists all agent profiles with names, descriptions, model, and allowed tools
`get_agent`	`name`, `skipCache?`	Fetches the full profile for a named agent
`list_docs`	none	Returns the documentation index as a hierarchical tree
`get_doc`	`path[]`, `skipCache?`	Fetches raw markdown source for a documentation page

How to Use

Prerequisites

A GitHub Fine-grained PAT with Contents: Read-only on Arda-cards/agentic-workspace and Arda-cards/documentation
The PAT encoded as URL-safe Base64

Encoding Your Token

echo -n "github_pat_YOUR_TOKEN" | base64 | tr '+/' '-_' | tr -d '='

If you use 1Password, store the encoded value alongside the original PAT:

ENCODED=$(op read "op://Private/GitHub Personal Access Token/token" \
  | tr -d '\n' | base64 | tr '+/' '-_' | tr -d '=') \
  && op item edit "GitHub Personal Access Token" --vault Private "base64-url-safe=$ENCODED"

Connecting from claude.ai

Open a claude.ai Project and go to Project Settings > MCP Servers > Add Server
Set:
- URL: https://claude-context-mcp.arda-cards.workers.dev/mcp
- Authentication: Bearer token
- Token: your URL-safe Base64-encoded token
Add the system prompt below to the Project

Connecting from Claude Code

With 1Password (recommended) — add to ~/.claude/settings.json under mcpServers:

{
  "claude-context": {
    "command": "/bin/sh",
    "args": [
      "-c",
      "TOKEN=$(op read 'op://Private/GitHub Personal Access Token/base64-url-safe') && npx mcp-remote https://claude-context-mcp.arda-cards.workers.dev/mcp --header \"Authorization: Bearer $TOKEN\""
    ]
  }
}

Without 1Password — add to ~/.claude/mcp.json:

{
  "mcpServers": {
    "claude-context": {
      "url": "https://claude-context-mcp.arda-cards.workers.dev/mcp",
      "headers": {
        "Authorization": "Bearer <your-URL-safe-Base64-encoded-token>"
      }
    }
  }
}

Recommended System Prompt

Add this to claude.ai Projects that should use the MCP server:

You have access to a Claude context MCP server with skills, agent profiles,
and project documentation.

For any task involving file creation or technical implementation (docx, pptx,
pdf, spreadsheets, code, frontend components, etc.): call list_skills, then
get_skill for the relevant skill and follow its instructions before proceeding.

When asked to take on a specialist role, or when a task would benefit from one:
call list_agents to see available profiles, then get_agent to load and adopt
the appropriate persona.

When you need context about system architecture, requirements, or design
decisions: call list_docs to see what documentation is available, then
get_doc with the relevant path array to fetch the content.

If content seems stale, use skipCache: true on get_* calls to fetch fresh.

Recommended Memory Nudge

Add to claude.ai memory so the server is used across all conversations:

I have a Claude context MCP server connected with skills, agent profiles, and
project documentation. For implementation tasks, always call list_skills and
get_skill before starting. When taking on a specialist role, call list_agents
and get_agent. When needing architecture or requirements context, call
list_docs and get_doc. Use skipCache: true if content seems outdated.

Index Freshness

The bundled indexes (list_* tools) are rebuilt on three triggers:

Trigger	When
Push to `main`	Any change to the MCP server code
Saturday 00:00 Pacific	Weekly rebuild picks up skill, agent, and doc changes from source repositories
Manual dispatch	On-demand via GitHub Actions workflow

A newly added skill or doc page will not appear in list_* results until the next rebuild. Use get_skill or get_doc directly if you know the name — those always fetch live from GitHub.

Source Repository

Arda-cards/skills-mcp — contains the Worker source, build scripts, tests, CI/CD, and the full README with operator setup instructions.

Working with AI Agent Assistants — how to collaborate with Claude Code agents, multi-agent teams, and session management
IDE and Tooling — development toolchain configuration