Enables Claude Code, Cursor, and Codex users to orchestrate plan/build/review workflows across host or CLI models with a spec gate, tool vetting, hard per-run and per-day cost/token limits, safe defaults, and clear logs.
Enables remote control of the Cursor AI agent via chat platforms like GitHub Issues or Telegram, allowing users to start sessions and receive summaries and prompts in a chat thread.
Enables Codex to rank available tools and skills, rerank code and documentation search results, and triage saved output artifacts using Jev-based relevance scoring while leaving final decisions to Codex.
Enables an orchestrator like Claude Code to hand coding work off to Cursor SDK agents, run as detached background jobs that survive the session and can be listed, inspected, steered, resumed, or stopped. Agents are asynchronous and reusable across follow-up turns, letting expensive frontier models delegate cheaply without blocking or paying to read the results.
An MCP relay that adds a use_jev tool to every connected server, letting the Jev model autonomously select and execute tool calls while the calling model supplies any required text values. It enables agents to delegate multi-step tool use across all configured MCP servers through a single unified interface.
Enables a supervisor agent to dispatch, monitor, follow up on, cancel, and inspect Cursor CLI code tasks through a governed MCP interface, with permission tiers, workspace allowlists, isolated git worktrees, and auditable artifacts. It also exposes an A2A-compatible HTTP endpoint for agent discovery and scheduling.
Enables Codex to manage a local Claude Code companion through MCP, implementing a dual-agent workflow where Codex handles reasoning and review while Claude Code performs engineering tasks.
Enables any MCP client to triage cloud agents, projects, conversations, transcripts, pull requests, and files, including listing agents, reading transcripts, following runs, and sending follow-ups, using an API key or browser sign-in.
Enables running a stateless Model Context Protocol server over Streamable HTTP on Cloudflare Workers or any fetch-capable runtime, with signed sessions, origin checks, and JSON-RPC request handling. It lets developers define tools with JSON Schema validation and per-request context.
Enables Claude Code to control Firefox through tools for tabs, navigation, clicks, typing, scrolling, screenshots, page reading, forms, JavaScript, and console/network logs.
Enables Codex to submit, poll, and retrieve bounded text handoffs for durable image-generation and image-inspection jobs while keeping provider-returned media bytes outside the MCP boundary.
Enables orchestrator models to delegate code creation and editing to Cursor's Composer 2 or Grok 4.5 agents, keeping the orchestrator's context free and shifting bulk code generation to cheaper agents.
Enables AI coding agents to retrieve the exact original tool-output slices omitted during lossless context compression, via on-demand expansion of hashed references stored in a shared cross-process SQLite vault. Tracks expansion overhead so agents and users can measure true net token savings across debugging sessions.
Enables Grok Bot to securely communicate with a local Cursor Agent CLI through an OAuth-authenticated MCP gateway, exposing exactly two tools for asking questions and checking status.
Enables AI agents to control Cursor IDE's built-in Browser Tab through a stdio MCP interface, providing accessibility snapshots with refs, click/type/fill, wait conditions, screenshots, and console/network inspection.
Automatically captures project state on git commits, detects context drift, and delivers structured token-budgeted briefings to AI agents, ensuring project continuity without manual effort.
Acts as a bridge between Claude's desktop application and the Cursor editor, enabling seamless AI-powered automation and multi-instance management across platforms with standardized communication and secure token-based authentication.
Exposes 26 browser_* MCP tools that let an agent drive a local Chrome instance—opening and focusing tabs, reading pages, clicking, typing, uploading files, taking screenshots, recording GIFs, and inspecting console and network activity—even when the client runs inside Docker. Control is scoped to a dedicated tab group and protected by token authentication.
MCP server enabling Cursor to delegate complex coding tasks to ChatGPT Web (GPT-5.6 Sol High) via Temporary Chats, supporting parallel batch reviews and up to five concurrent sessions.
Enables users to record Windows desktop and browser workflows once and generate replayable Cursor Skills, using Playwright MCP and Windows Computer MCP for semantic, non-coordinate replay.
Enables Codex to run image generation and inspection as durable, restart-reconcilable local jobs that keep provider-returned media behind the tool boundary and hand back only bounded text, hashes, relative references, and Job IDs. It also supports idempotent submission, job history, and an offline deterministic mock provider for validating transport and persistence behavior without spending API credits.
Enables users to describe an LLM fine-tuning workload once and receive ranked, task-aware GPU recommendations across a dozen cloud providers based on required VRAM/FLOPs, cost, and runtime. It surfaces the cheapest, fastest, and balanced options through MCP so coding assistants can plan training hardware without manual price comparison.
Local Context Window Profiler, KV-Cache TTL Monitor, Multi-Family FTS5 Output Compressor, and Deterministic PreInvocation/PreToolUse Surgeon for Claude Code, Google Antigravity, Codex CLI, Gemini CLI, and Cursor.
Connects Windows-based Codex to the Cursor Agent CLI in WSL, providing a run_cursor tool to execute tasks in specified working directories and resume sessions per directory.
An MCP server that keeps raw tool output out of the context window by sandboxing and compressing results (about 98% reduction), persisting file edits, git operations, tasks and decisions in SQLite with BM25 retrieval so the agent resumes after compaction, and enforcing routing across 17 platforms via MCP and hooks. It also encourages the model to write scripts that compute answers and log only results instead of reading large data into context.
MCP server providing lossless file reads with deduplication, per-repo token metering, and HMAC-signed context receipts. Enables auditable, vendor-neutral measurement of what an AI agent saw.
Provides a single MCP gateway that discovers, plans, lazy-loads, calls, and unloads tools from downstream MCP servers on demand, reducing context usage while keeping all tools accessible through one connection.
Enables AI assistants and other MCP clients to launch and steer Cursor Cloud Agents on repositories or as repo-less research tasks, including model selection, mid-run follow-ups, cancellation, result retrieval, and run listing. Also exposes authentication checks and per-agent or per-run token usage and cost reporting.
Enables multiplexing multiple MCP servers behind a single gateway connection that lazily loads only the tool schemas needed for the current turn, cutting context usage while providing OAuth authentication, circuit breakers, and hot reload.
A local plugin for Codex and Claude Code that replaces earlier images with short placeholders. Its MCP tool retrieves the original image by ID when needed, even after context compaction.
A reusable MCP server providing shared, versioned context across AI agents and devices via a private GitHub workspace, with tools to discover projects, bootstrap, query, and close out task state.
Aggregator MCP proxy that collapses N downstream MCP servers into 4 meta-tools with progressive tool discovery, and compresses large tool outputs (HTML→Markdown, JSON summarization) with full-output retrieval via read_more and a per-session token-savings report.
Enables Cursor to reach any number of child MCP servers through just five host tools and a staged catalog, so tool schemas stay off the host until a specific tool is needed and token spend stays low. Searches tool names, pulls a single schema on demand, then invokes it, while disabled children consume zero processes and zero schemas.
Proxy for OpenAI-compatible backends that compacts conversation history before the context window fills, with read-only MCP tools to search and read back the archived originals.
Enables agents to load instruction modules on demand from linked git repos, keeping the always-loaded AGENTS.md slim. It syncs each module's repo before reading and returns the module text with the commit it was read at.
Provides persistent memory, identity, and context for Claude sessions through a memory graph, enabling recall, storage, and relationship management of entities and decisions.
Enables multiple AI coding agents and humans to coordinate through a Git-native Markdown blackboard (.context/ + AGENTS.md), where they can inspect workspace overviews, claim and toggle ready tasks, record architecture decisions, synthesize shared prompts, and hand off work via atomic, lock-protected writes that prevent lost updates and formatting drift. It also serves a fully offline local web dashboard that live-syncs any blackboard change over WebSocket.
Enables coding agents to run budgeted tests and checks under a countdown with a stop rule, shape long logs into concise failure summaries, and guard pre-existing test files against silent assertion rewrites. It also delivers relevance-filtered project memory and renders a token-and-time roadmap of the orchestrator and every subagent spawned for a task.
Exposes a context_folding MCP tool that lets an agent mark conversation history entries (user messages, model replies, tool results) as foldable, keep, or unmarked and replace marked content in outgoing model requests with a head, a concise summary, and an original-entry ID, without altering the client's own full history. Enables listing, paginated reading of originals, layered expansion via previous IDs down to the original, status inspection, and cleanup, with folding applied only when token thresholds and trusted cost-accounting gates are satisfied.
Enables ZCode to control a real Chrome browser, including tabs, navigation, page content, clicks, form filling, keyboard input, scrolling, screenshots, and JavaScript evaluation through a local MCP bridge.
This MCP server provides tools to manage, score, compress, and prune AI agent conversation context, helping keep agents focused and reduce token costs. It is a free, local, pure Python solution for any MCP client.
Enables selecting exact conversation excerpts, carrying forward only confirmed background and provenance, and handing the packaged selection to another agent for continuation.
Enables high-volume MCP resource queries to be paginated with cryptographically signed, stateful cursors, automatic TTL cleanup, and bounded memory use so agent clients avoid context-window overflows and timeouts.
Enables MCP clients to interact with A2A agents through four ordinary tools, translating agent discovery, messaging, task reading, and cancellation between the A2A and MCP protocols, with support for the tasks extension.
Reduces token consumption for AI coding agents by 50-70% through intelligent code context filtering, Git delta tracking, and local SQLite/Tree-sitter indexing.
Enables Alexa+ to retain and recall a lightweight per-user context thread across separate voice interactions, supporting starting, querying, and listing active contexts for continuity.
Enables AI agents to securely retrieve pre-compiled bounded context and memory state over a zero-network Unix Domain Socket, ensuring tenant isolation and sub-millisecond rehydration in Kubernetes environments.
Enables users to compress text, extract Python code signatures, hydrate saved session state, and save checkpoints, reducing token usage when handing context between models and agents.
Acts as a bridge between standard REST/OpenAPI web APIs and the Model Context Protocol, dynamically converting REST endpoints into MCP tools for LLM clients like Cursor, Claude Desktop, and n8n.
Self-hosted, single-owner MCP relay that lets an agent client queue requests for an authorized ChatGPT worker to look up context from permitted sources, apply bounded edits to a specific Page, or save a selected transcript as a new Page. Requests, results, and status are stored in the owner's own Cloudflare account, with the worker woken by signed events and results retrieved by polling.
Enables a person to point their own AI agent at specific on-screen app elements over MCP, letting the agent read the underlying data without sharing the screen or receiving visible content.