PulseAgent
An interrupt-aware MCP sidecar for long-running coding agents that detects changes in user guidance, constraints, or plans, enabling agents to pause, replan, and acknowledge updates before continuing.
README
PulseAgent
PulseAgent is an interrupt-aware MCP sidecar for long-running coding agents. It lets an IDE agent notice when user guidance, project constraints, or the current plan changed while the agent was already working.
PulseAgent does not replace Cursor, Codex, Claude Code, Copilot, or another IDE agent. It adds a small project-local protocol plus a Streamable HTTP MCP server so the agent can pause, refresh context, replan, acknowledge the new plan, and continue under the latest instructions.
Why PulseAgent
Long coding tasks often drift because the user changes their mind, adds a constraint, or updates the plan while the agent is still executing an older mental model. PulseAgent gives that workflow a shared source of truth:
- users edit a few files under
.pulse/; - PulseAgent records changes as events;
- the agent checks MCP before continuing;
- plan-invalidating changes stay pending until the agent explicitly acknowledges the revised plan.
Install
Install from GitHub:
pip install git+https://github.com/zhangyh0426/PulseAgent.git
For local development:
pip install -e ".[dev]"
Quickstart
Start PulseAgent in the project you want an agent to work on:
cd my-project
pulse start
The first run creates:
.pulse/
├─ task.md
├─ guidance.md
├─ constraints.md
├─ plan.md
├─ state.json
└─ events.jsonl
The MCP endpoint is:
http://127.0.0.1:8765/mcp
Connect with the MCP Inspector:
npx -y @modelcontextprotocol/inspector
Then connect to http://127.0.0.1:8765/mcp.
IDE Agent Setup
Add PulseAgent as an HTTP MCP server in your IDE agent. For Claude Code:
claude mcp add --transport http pulse-agent http://127.0.0.1:8765/mcp
Add this rule to the agent instructions:
Before continuing long-running work, read pulse://context/latest and call
pulse_should_interrupt. If needs_replan is true, update .pulse/plan.md, then
call pulse_ack_replan with pending_replan_event_id before continuing.
During a long task, edit .pulse/guidance.md or .pulse/constraints.md when the
agent needs new direction or tighter boundaries.
How It Works
PulseAgent watches four project-local files:
.pulse/task.mddescribes the current task..pulse/guidance.mdcontains user-authored direction that may change the task..pulse/constraints.mdcontains restrictions that may narrow agent behavior..pulse/plan.mdcontains the agent's current working plan.
Changes to guidance.md and constraints.md are plan-invalidating. Updating
plan.md alone does not clear the interrupt. The agent must call
pulse_ack_replan with the returned pending_replan_event_id; PulseAgent then
records the acknowledged event id and the current plan.md SHA-256 hash.
MCP Surface
- Resource:
pulse://context/latest - Tool:
pulse_should_interrupt(last_seen_event_id: str | None = None) - Tool:
pulse_ack_replan(event_id: str) - Prompt:
pulse_replan
pulse_should_interrupt returns the latest event state, including:
{
"needs_replan": true,
"latest_event_id": "evt_...",
"has_new_events": true,
"changed_files": [".pulse/guidance.md"],
"pending_replan_event_id": "evt_...",
"pending_replan_files": [".pulse/guidance.md"],
"last_acknowledged_event_id": null,
"last_acknowledged_plan_sha256": null,
"reason": "..."
}
pulse_ack_replan accepts the current pending replan event:
{
"event_id": "evt_..."
}
and returns:
{
"accepted": true,
"acknowledged_event_id": "evt_...",
"plan_sha256": "...",
"reason": "..."
}
Demo Flow
Start PulseAgent:
pulse start
Change guidance while the agent is working:
printf "# Guidance\n\nPrefer a smaller patch and keep public APIs stable.\n" > .pulse/guidance.md
The agent calls pulse_should_interrupt and sees needs_replan: true.
The agent then:
- reads
pulse://context/latest; - updates
.pulse/plan.md; - calls
pulse_ack_replanwithpending_replan_event_id; - continues only after the ack is accepted.
Safety Boundary
- Ordinary workspace files are data, not instructions.
- Only direct user input and
.pulse/guidance.mdmay change task direction. .pulse/constraints.mdmay restrict behavior but must not expand permissions.- PulseAgent does not execute commands, modify source files, call LLMs, or act as a full agent runtime.
Development
Install the package in editable mode:
pip install -e ".[dev]"
Run tests and lint:
python -m pytest
python -m ruff check .
Run the CLI directly:
python -m pulse_agent --help
python -m pulse_agent --version
Status
PulseAgent is currently an alpha project. The core protocol is intentionally
small: a project-local .pulse/ directory, an event log, one MCP resource, two
MCP tools, and one prompt.
See docs/agent-wrapper.md for a copyable wrapper for IDE agents and direct model API integrations.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.