mcp-resource-subscribe-test
Minimal MCP server for testing client support for resources/subscribe and notifications/resources/updated, simulating an asynchronous status update.
README
mcp-resource-subscriber
CLI probe for MCP resources/subscribe — connects to any MCP Streamable HTTP server, subscribes to a resource, receives the live update notification, and re-reads the updated content.
Install
# recommended (no install):
pnpm dlx mcp-resource-subscriber --url <mcp-server-url> --uri <resource-uri>
# fallback if pnpm is unavailable:
npx mcp-resource-subscriber --url <mcp-server-url> --uri <resource-uri>
# or install globally:
npm install -g mcp-resource-subscriber
mcp-resource-subscriber --url <mcp-server-url> --uri <resource-uri>
Note: A reference MCP test server used during compatibility verification is also included in this repository (Docker Compose). See the Lab Server section below.
CLI Usage
Against copilot-review-mcp
mcp-resource-subscriber \
--url http://127.0.0.1:8080/mcp/copilot-review \
--uri copilot-review://watch/<watch_id> \
--timeout-ms 900000
copilot-review-mcp is the @scottlz0310/copilot-review-mcp server. Replace <watch_id> with the ID returned by start_copilot_review_watch.
Against the bundled test server
# Start the test server first:
docker compose up --build
# or: npm run dev
mcp-resource-subscriber --url http://127.0.0.1:8089/mcp
Note:
test://review/statusis the default resource URI and is only meaningful against the bundled test server. For any other MCP server, always pass--uriexplicitly.
Options
--url <url> MCP server Streamable HTTP endpoint (required)
Env: MCP_PROBE_URL
--uri <uri> Resource URI to subscribe to
Default: test://review/status (bundled test server only)
Env: MCP_PROBE_URI
--auth-token <tok> Bearer token for Authorization header
Prefer MCP_PROBE_AUTH_TOKEN env var (flag is visible in
process lists and may be stored in shell history)
Env: MCP_PROBE_AUTH_TOKEN (recommended)
--skip-resource-list-check
Skip resources/list and assume the URI exists.
Use for servers with dynamic resources not in list.
Env: MCP_PROBE_SKIP_LIST_CHECK=true
--timeout-ms <ms> Notification wait timeout in ms (default: 15000)
Env: MCP_PROBE_TIMEOUT_MS
--version, -v Print version and exit
--help, -h Print this help and exit
Structured output
Every run emits machine-parseable lines:
capabilities {"subscribe":true,"listChanged":true}
resource-found true
resource-uri <resource-uri>
server-url <url>
initial
<initial resource text>
route subscription
subscribed true
notification-received true
unsubscribed true
recommended_next_action READ_REVIEW_THREADS
error-code null
notification <resource-uri>
final
<updated resource text>
phase-summary route=subscription url=<url> uri=<uri>
Note:
recommended_next_actionis only emitted when the final resource text contains it (e.g., fromcopilot-review-mcp). It is omitted for the bundled test server.
On failure:
error-code SERVER_URL_UNKNOWN
phase-summary route=failed url=unknown error-code=SERVER_URL_UNKNOWN
error-code RESOURCE_NOT_FOUND
phase-summary route=timeout url=<url> uri=<uri> error-code=RESOURCE_NOT_FOUND
error-code NOTIFICATION_TIMEOUT
phase-summary route=timeout url=<url> uri=<uri> error-code=NOTIFICATION_TIMEOUT
Lab Server
Minimal MCP Streamable HTTP server for testing whether MCP clients correctly handle resources/subscribe and notifications/resources/updated.
This repository is meant to be a reproducible issue / compatibility lab for CLI AI agents such as Codex CLI, Gemini CLI, OpenCode, GitHub Copilot CLI, Claude Code, Goose, and Crush.
Purpose
The server exposes one fixed MCP resource:
test://review/status
Initial content:
status: pending
version: 1
message: Waiting for simulated review result.
After a client subscribes to the resource, the server waits for MCP_TEST_UPDATE_DELAY_SECONDS, changes the resource, and sends:
{
"method": "notifications/resources/updated",
"params": {
"uri": "test://review/status"
}
}
Updated content:
status: reviewed
version: 2
message: Simulated review result is now available.
Why Resources/Subscribe Instead Of Tools/Call
tools/call is useful for explicit actions, but many agent workflows depend on context that changes after the original request. Polling every source is noisy and client-specific. MCP resource subscriptions give clients a protocol-level way to learn that a known context object changed and should be re-read.
Examples where subscription behavior matters:
- Copilot review result
- PR review thread
- CI status
- Codecov comment
- GitHub issue discussion
- local build/test result
This test server focuses on whether the client notices a resource update, re-runs resources/read, and reflects the new content in the agent loop / model context.
Start
docker compose up --build
MCP URL:
http://127.0.0.1:8089/mcp
For local development:
npm install
npm run dev
Configuration
| Environment variable | Default | Description |
|---|---|---|
MCP_TEST_PORT |
8089 |
TCP port the server listens on |
MCP_TEST_PATH |
/mcp |
Additional MCP endpoint path. The server always registers /mcp; this adds a second path (e.g. /mcp/subscribe-probe for gateway routing). Both paths share the same MCP handler. |
MCP_TEST_UPDATE_DELAY_SECONDS |
5 |
Seconds to wait before sending the resource update notification |
MCP_TEST_INITIAL_STATUS |
pending |
Initial value of the status field in the resource |
MCP_TEST_UPDATED_STATUS |
reviewed |
Value of status after the simulated update |
MCP_TEST_SEND_LIST_CHANGED |
false |
Also send notifications/resources/list_changed after the update |
MCP_TEST_LOG_LEVEL |
debug |
Log verbosity (debug / info / warn / error / silent) |
If MCP_TEST_SEND_LIST_CHANGED=true, the server also sends notifications/resources/list_changed after the simulated update.
Expected Client Behavior
An ideal MCP client should follow this flow:
initialize
↓
resources/list
↓
resources/read test://review/status
↓
resources/subscribe test://review/status
↓
receive notifications/resources/updated
↓
resources/read test://review/status again
↓
reflect updated status: reviewed in agent context
Server Capabilities
The initialize response advertises:
{
"resources": {
"subscribe": true,
"listChanged": true
}
}
Implemented MCP Messages
initializeresources/listresources/readresources/subscriberesources/unsubscribenotifications/resources/updatednotifications/resources/list_changedwhenMCP_TEST_SEND_LIST_CHANGED=true
No tools are implemented.
Logs
The server logs each important message so client behavior can be checked objectively:
[initialize] client connected
[resources/list] requested
[resources/read] uri=test://review/status version=1
[resources/subscribe] uri=test://review/status
[resource/update] uri=test://review/status version=2
[notification/send] notifications/resources/updated uri=test://review/status
[resources/read] uri=test://review/status version=2
[resources/unsubscribe] uri=test://review/status
The key evidence for resource subscription support is:
resources/subscribe was received
notification was sent
resources/read was received again after the notification
Tests
npm test
The test suite verifies:
resources/listreturnstest://review/status- initial
resources/readreturns version 1 resources/subscribetriggers an internal update to version 2notifications/resources/updatedis received- updated
resources/readreturns version 2
Standalone Subscription Probe Client
The repository also includes a reusable MCP SDK client that exercises the full subscription flow against a running server:
npm run probe:subscribe -- --url http://127.0.0.1:8089/mcp
After npm run build, the same client can be run directly with Node:
node dist/src/client/cli.js --url http://127.0.0.1:8089/mcp
This client is separate from any AI client's native MCP surface. For Codex CLI, it demonstrates a reproducible agent-driven SDK workaround: if the agent has shell, Node.js, local dependency, and localhost network access, it can run this client to call resources/subscribe, receive notifications/resources/updated, and re-read the updated resource.
Verification Procedure
Use docs/verification-guide.md for a repeatable client verification procedure.
Record results in results/compatibility-matrix.md.
Skill Templates
Reusable Codex skill templates are tracked under docs/skills. The pr-review-subscribe template documents a PR review cycle that uses MCP resources/subscribe as the primary wait route and polling only as fallback.
Client Compatibility
See results/compatibility-matrix-v2.md for the current Round 2 compatibility matrix (tool + resource testing) across Codex CLI, Gemini CLI, OpenCode, GitHub Copilot CLI, Claude Code, Goose, and Crush.
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
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
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.