cursor-bridge-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@cursor-bridge-mcp Show me the active rules for my current files"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Cursor Bridge MCP
A local MCP server that gives Claude Code (and any MCP-compatible AI tool) native understanding of Cursor's context ecosystem — rules, agents, and skills.
Why
Most teams aren't all-in on one AI tool. Some developers prefer Cursor, others use Claude Code, and some switch between both depending on the task. But Cursor's context system — rules, agents, and skills — only works inside Cursor. Claude Code ignores .cursor/ entirely and reads only CLAUDE.md.
That leaves teams with a choice: maintain parallel context systems that drift apart, or pick one tool and lose the other's strengths. Cursor Bridge eliminates this by making Cursor's context files consumable by any MCP-compatible AI tool. Write your conventions once in .cursor/rules/, define your agents in .cursor/agents/, package your skills in .cursor/skills/ — and every developer on the team gets the same context regardless of which tool they use.
Related MCP server: Context7 MCP Server
What It Does
Cursor Bridge reads .cursor/rules/, .cursor/agents/, and .cursor/skills/, applies Cursor's activation logic (glob matching, always-apply, description-based discovery), and serves the right context to any AI tool on demand.
Rules are routed intelligently — always-apply rules inject automatically, glob-matched rules activate based on the files you're working on, and description-matched rules surface when the task is relevant
Agents are served as loadable role instructions — when a task matches an agent's description, the AI can adopt that agent's behavioral guidance
Skills are served as on-demand knowledge packages — the AI browses the index and loads what it needs, including supporting scripts and references
Quick Start
The fastest way to get running — no install needed:
{
"mcpServers": {
"cursor-bridge": {
"command": "npx",
"args": ["cursor-bridge-mcp"]
}
}
}Add this to .claude/settings.json (Claude Code) or .cursor/mcp.json (Cursor). That's it — npx downloads and runs the latest version automatically.
Other Install Options
Install globally
npm install -g cursor-bridge-mcpThen use "command": "cursor-bridge-mcp" in your MCP config instead of npx.
Build from source
git clone https://github.com/theamazingwolf/cursor-bridge-mcp.git
cd cursor-bridge-mcp
npm install
npm run build
npm linkThen use "command": "cursor-bridge-mcp" in your MCP config, or point directly to the build:
{
"mcpServers": {
"cursor-bridge": {
"command": "node",
"args": ["/path/to/cursor-bridge-mcp/dist/index.cjs"]
}
}
}Tools
get_active_rules
Returns rules that apply to your current files and/or task. This is the primary tool — call it when starting work or switching file context.
{
"files": ["src/components/Button.tsx"],
"task_description": "Refactoring the button component"
}Rules are matched by activation mode:
always —
alwaysApply: truerules are always includedauto — rules with
globspatterns are included when files matchagent-requested — rules with no globs are included when the task description overlaps with the rule's description
get_context_index
Returns a lightweight menu of all available rules, agents, and skills — descriptions only, no content. Use this to discover what's available.
{
"types": ["rule", "agent", "skill"]
}get_rule
Fetches the full content of a specific rule by name.
{ "name": "data-models-relationships" }get_agent
Fetches the full definition of a Cursor agent — including its system prompt / instructions.
{ "name": "security-auditor" }get_skill
Fetches the full content of a skill package.
{
"name": "pdf-processing",
"include_supporting_files": true
}get_skill_file
Fetches a specific supporting file from within a skill directory.
{
"skill_name": "pdf-processing",
"file_path": "scripts/extract-text.py"
}Usage with Claude Code
Add a CLAUDE.md to any project with a .cursor/ directory to tell Claude Code to use the bridge automatically. Here's a template:
## Cursor Bridge MCP
When the cursor-bridge MCP is connected, use it to read Cursor context files
instead of reading .cursor/ files directly.
- On session start, call `get_context_index` to discover available context
- When working on files, call `get_active_rules` with current file paths and/or
a task description to get relevant rules with full content
- Use `get_rule`, `get_agent`, `get_skill` to load specific items by name
- Use `get_skill_file` to load supporting files from skill packages
The bridge applies Cursor's activation logic (glob matching, always-apply,
description relevance) so you get the right context automatically.This way Claude Code knows to call the MCP tools rather than manually reading and interpreting .cursor/ files. The bridge handles the routing logic — which rules apply, which don't — so the AI doesn't have to.
Cursor Context Formats
Rules (.cursor/rules/)
Standard .mdc or RULE.md files with YAML frontmatter:
---
description: "React component conventions"
globs: "src/components/**/*.tsx"
alwaysApply: false
---
# Component Patterns
Your conventions here...Both .mdc files directly in the rules directory and RULE.md files in subdirectories are supported.
Agents (.cursor/agents/)
Markdown files with agent metadata in frontmatter:
---
description: "Reviews code for security vulnerabilities"
model: inherit
readonly: true
is_background: true
---
You are a security specialist...Skills (.cursor/skills/)
Directory-based packages with a SKILL.md and optional scripts/, references/, assets/ subdirectories:
.cursor/skills/pdf-processing/
├── SKILL.md
├── scripts/extract-text.py
└── references/form-fields.mdConfiguration
Optionally create .cursor-bridge/config.yaml to customize directory paths:
rules_directory: ".cursor/rules"
agents_directory: ".cursor/agents"
skills_directory: ".cursor/skills"All paths are relative to the workspace root. If no config file exists, the defaults above are used.
How It Works
No file watchers — the index rebuilds from disk on each tool call, parsing only YAML frontmatter (not full file content). This keeps every call fresh and avoids platform-specific reliability issues.
Lazy content loading — full markdown bodies are only read when a tool specifically needs them (e.g.,
get_rule,get_active_rulesfor matched rules). The index itself never loads content.No network access — pure local file parsing. No AI calls, no external APIs.
No code execution — skill scripts are served as text content, never executed.
Development
npm install
npm run build
npm testAvailable Tools
6 toolsget_active_rulesA
Returns rules that apply to a given set of files and/or task description. Call at session start or when switching file context.
Inputs: files, task_description.
get_agentB
Fetches the full definition of a Cursor agent by name.
Inputs: name (required).
get_context_indexA
Returns a lightweight menu of all available context — rules, agents, and skills — with descriptions only, no content.
Inputs: types.
get_ruleA
Fetches the full content of a specific rule by name.
Inputs: name (required).
get_skillC
Fetches the full content of a skill package.
Inputs: name (required), include_supporting_files.
get_skill_fileA
Fetches a specific supporting file from within a skill directory.
Inputs: file_path (required), skill_name (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool targets a distinct entity (rules, agents, skills, context index) with clear parameters. There is no overlap: get_active_rules and get_rule differ by specificity, and get_skill and get_skill_file differ by scope.
All tools follow a consistent 'get_' prefix followed by the entity name in snake_case. This uniform pattern makes it easy to predict tool names.
With 6 tools, the set is appropriately scoped for a read-only retrieval server. Each tool serves a necessary function without redundancy or bloat.
The tool set covers all primary retrieval operations for the domain: fetching active rules, individual rules, agents, skills, skill files, and listing all available context. No obvious gaps for a read-only interface.
Maintenance
Related MCP Connectors
One shared context your team's AI tools read & write over MCP. No re-explaining. Free.
Your memory, everywhere AI goes. Build knowledge once, access it via MCP anywhere.
Shared brain, one current version of memory for any MCP client: ChatGPT, Claude, Gemini, Cursor
Share one project context across ChatGPT, Claude, Telegram and any MCP client.
Related MCP Servers
- AlicenseNot gradedqualityNot gradedmaintenanceEnables programmatic control of Cursor's cloud-based AI agents for automated code generation and repository management via standardized MCP tools.27 npm6MIT
- AlicenseNot gradedqualityDmaintenanceEnables LLM-powered code analysis, generation, debugging, and context management through MCP integration with IDEs like Cursor and Claude Desktop.MIT
- FlicenseNot gradedqualityDmaintenanceEnables access to PolyAI Agent Studio Cursor rules via MCP tools for listing, retrieving, and searching rules, making them usable by any MCP-compatible client.-
- AlicenseNot gradedqualityBmaintenanceEnables AI coding agents to retrieve repository context through stdio-based MCP tools, including ranked repo maps, context packs, and live symbol search.6 npmMIT