M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
K

Kubeshark

Official
Kubernetes network observability for SREs and AI agents
How to install ↓ Compare GitHub source repository ↗
Category
Other #131 of 230
Stars
★ 12.1k Very popular
Transport
stdio (local process)
Runtime
Prebuilt binary · Go
Credentials
No credential needed
License
Apache-2.0
Last commit
Tools
14
52FMRS · D

Kubeshark is an official MCP server maintained by Kubeshark for Kubernetes network observability, debugging, incident response, and AI-assisted root-cause analysis. It uses stdio and supports direct URL and kubeconfig proxy modes. Key considerations are cluster access, destructive deployment and removal operations, and sensitive-data exposure from decrypted traffic and full payloads.

Strongest · Setup experience 12/20 Weakest · Security and permissions 8/20

Reliability
9/20
Security and permissions
8/20
Maintenance
11/20
Documentation
12/20
Setup experience
12/20
Why each score
Reliability 9/20
The manifest declares stdio packages, proxy modes, and a coherent tool set, making the normal path plausible from static evidence. However, versions, download URLs, and hashes are generated placeholders, and the supplied material contains no CI, committed tests, initialization-handshake evidence, or diagnosable error behavior. Points are deducted, and the static-review ceiling of 12 is respected.
Security and permissions 8/20
Destructive start and stop operations are disabled by default and require an explicit allow-destructive setting, which is positive permission evidence. However, the server can access cluster-wide traffic, headers, bodies, PCAPs, and decrypted TLS data; the manifest does not document authentication, authorization, data scoping, sensitive-data protection, or access controls for direct URLs. This leaves substantial data-boundary and least-privilege gaps, but no stated red-line condition is established.
Maintenance 11/20
The repository is marked unarchived and identifies Kubeshark as the maintainer, provides an Apache-2.0 license, release access, and contribution guidance. These are useful maintenance signals. The supplied sources do not establish commit cadence, release history, dependency-update practice, vulnerability response, or a security contact, so a high score is not justified.
Documentation 12/20
The README provides Helm, Homebrew, and binary installation paths, MCP connection instructions, capability descriptions, example questions, protocol coverage, and links to supporting documentation. It does not provide per-tool parameters and limits, authentication details, resource costs, troubleshooting, permission boundaries, or guidance for handling sensitive traffic, so completeness is reduced.
Setup experience 12/20
The main path from installing via Helm or Homebrew to connecting through kubectl port-forward is reasonably clear, and the manifest describes URL and proxy modes plus several platforms. Use still depends on Kubernetes, kubectl, kubeconfig access, and an existing deployment; package URLs contain X.Y.Z placeholders, and no complete mainstream-client configuration or verification flow is supplied. The score therefore remains moderate and below the static ceiling.

Static review · not runListed 2026-07-29

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • SRE, DevOps, and network engineering teams using Kubernetes.
  • Incident response and root-cause analysis requiring cluster-wide network, API, and Kubernetes context.
  • Teams that want MCP-compatible AI agents to query live or retained network data.

Not for

  • Users who do not run Kubernetes or do not need network traffic observability.
  • Client scenarios requiring only generic file, database, or collaboration features.
  • Environments where handling sensitive traffic or decrypting TLS/mTLS is unacceptable.

Required permissions

  • Proxy mode requires a kubeconfig with access to the Kubernetes cluster.
  • Direct URL mode requires access to an existing Kubeshark Hub URL.
  • Start and stop operations require explicitly enabling --allow-destructive.
  • The source does not specify exact Kubernetes RBAC permissions or authentication mechanisms.

Risks and side effects

  • The server can deploy or remove Kubeshark; start_kubeshark and stop_kubeshark are marked destructive.
  • Observed data may include complete API requests, headers, bodies, and PCAPs.
  • TLS/mTLS traffic can be decrypted into plaintext, potentially exposing sensitive information.
  • The source does not specify access-control, authentication, or data-retention policies.

Setup

Before you start

Runtime：Prebuilt binary · Go

Install and deploy Kubeshark with: helm repo add kubeshark https://helm.kubeshark.com && helm install kubeshark kubeshark/kubeshark. Then optionally run kubectl port-forward svc/kubeshark-front 8899:80 and open http://localhost:8899. To connect an AI agent, run brew install kubeshark followed by claude mcp add kubeshark -- kubeshark mcp. You can also connect directly to an existing Kubeshark Hub with the url setting, or use proxy mode with a kubeconfig file.

Check that it works

After installing with brew install kubeshark and running claude mcp add kubeshark -- kubeshark mcp, tools like check_kubeshark_status should appear in the client's tool list; invoking it should return Kubeshark's running status in the cluster.

Troubleshooting

  1. Call check_kubeshark_status first to confirm that Kubeshark is running in the cluster.
  2. In proxy mode, verify the kubeconfig path and cluster access.
  3. In direct mode, verify that the Kubeshark Hub URL is reachable.
  4. For the local dashboard, confirm that kubectl port-forward svc/kubeshark-front 8899:80 is running.
  5. Check get_dissection_status and enable or disable L7 dissection as needed.
  6. Confirm that --allow-destructive has not been enabled without authorization.

Things to try

Once connected, you can ask your AI assistant things like:

  • Check whether Kubeshark is currently running in my cluster
  • List workloads with recent L7 traffic
  • Why did checkout fail at 2:15 PM? Find the related API errors
  • Create a PCAP snapshot of the current traffic

Tools 14

check_kubeshark_status read-only
Check whether Kubeshark is currently running in the cluster; read-only.
start_kubeshark destructive
Deploy Kubeshark to the Kubernetes cluster; requires --allow-destructive.
stop_kubeshark destructive
Remove Kubeshark from the Kubernetes cluster; requires --allow-destructive.
list_workloads read-only
List pods, services, namespaces, and nodes with observed L7 traffic.
list_api_calls read-only
Query L7 API transactions for HTTP, gRPC, Redis, Kafka, and DNS using KFL filters.
get_api_call read-only
Get detailed information about a specific API call, including headers and body.
get_api_stats read-only
Get aggregated API statistics and metrics.
list_l4_flows read-only
List L4 TCP/UDP network flows with traffic statistics.
Show 6 more tools
get_l4_flow_summary read-only
Get an L4 connectivity summary, including top talkers and cross-namespace traffic.
list_snapshots read-only
List all PCAP snapshots.
create_snapshot writes
Create a new PCAP snapshot of captured traffic.
get_dissection_status read-only
Check L7 protocol parsing status.
enable_dissection writes
Enable L7 protocol dissection.
disable_dissection writes
Disable L7 protocol dissection.

Use cases

Query API traffic and analyze errors, failures, and latency.
Trace requests through microservices.
View service communication topology and workload dependencies.
Analyze TCP/UDP connectivity, cross-namespace traffic, and top talkers.
Create or query PCAP snapshots for incident response and root-cause analysis.
Audit traffic for security concerns.

Supported clients

Claude Code
CursorPartial support

Listed from the project's documentation, not tested by this site.

Overview

Kubeshark is an eBPF-powered MCP server for Kubernetes network observability. It indexes cluster-wide L4/L7 traffic with Kubernetes, API, and network semantics, and analyzes protocols including HTTP, gRPC, Redis, Kafka, and DNS. It supports querying API calls, network flows, metrics, and PCAP snapshots, and can decrypt TLS/mTLS traffic without key management or sidecars.

Similar servers

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 659 · Tools 11 Compare with this →
Compare with other MCP servers

Source revision e27687a8ed41 Data synced 2026-10-19 Read the FMRS scoring method