Skip to main content
Glama

pipelines-mcp

Read-only MCP server for pipeline jobs, pods, configs, and logs.

It talks to the dev EKS cluster (pipelines-prd namespace) and the log store. Region us-east-2. SSO profile reveliolabs is only for EKS.

Run

Needs Python 3.13 and uv. Cluster tools need AWS SSO for the reveliolabs profile.

uv sync
uv run pipelines-mcp

The process starts without talking to AWS. Cluster tools log you in if needed. A helper on 127.0.0.1:18201 gets the login; retry. If none, they return a URL. Over SSH: RemoteForward 18201 127.0.0.1:18201.

Related MCP server: Claude Ops Investigator

Use in OpenCode

Add this to ~/.config/opencode/opencode.json, or to an opencode.json in a project. Use the real absolute path to this repo.

{
  "mcp": {
    "pipelines": {
      "type": "local",
      "command": [
        "uv",
        "run",
        "--directory",
        "/absolute/path/to/pipelines-mcp",
        "pipelines-mcp"
      ],
      "enabled": true
    }
  }
}

uv must be on PATH. Restart OpenCode after saving.

If the user gives a GitHub or Jenkins run, read that CI log first for the request_id UUID (not the run id), then read pipeline logs. If a cluster tool says AWS login started, retry the same request. Do not ask the user to open a URL unless the tool returned one.

Dev

uv sync --group dev
uv run pytest
uv run ruff check src tests
uv run basedpyright

Available Tools

8 tools
get_pipeline_jobB

Get one job by request_id, step_index, and replica.

Inputs: replica (required), request_id (required), step_index (required). Returns: One cluster job. Spec is never a field.

get_pipeline_job_configA

Get the full YAML config for one job by request_id, step_index, and replica. Secrets are redacted.

Inputs: replica (required), request_id (required), step_index (required). Returns: Redacted YAML config for one job or pod.

get_pipeline_logA

Read pipeline worker logs for a request. Empty until the worker pod is running. Default is the last 100 lines plus a cursor. Pass cursor to get only new lines. full=true still caps size.

Inputs: request_id (required), full, cursor. Returns: Capped log page with an opaque cursor.

get_pipeline_podA

Get one pod by pod_name.

Inputs: pod_name (required). Returns: One cluster pod.

get_pipeline_pod_configA

Get the full YAML config for one pod by pod_name. Secrets are redacted.

Inputs: pod_name (required). Returns: Redacted YAML config for one job or pod.

get_pipeline_service_logB

Read service logs for a request. Same cursor rules as pipeline log.

Inputs: request_id (required), full, cursor. Returns: Capped log page with an opaque cursor.

list_pipeline_jobsA

List pipeline jobs, newest first. Optional filters: status (Active, Complete, Failed, Pending, or running), request_id, limit (default 50, max 100). Job names are pipelines-{request_id}-{step_index}-{replica}.

Inputs: limit, status, request_id. Returns: result

list_pipeline_podsB

List pods for one job. Empty if they are already gone.

Inputs: replica (required), request_id (required), step_index (required). Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.8/5.0

    Scored across 8 tools

    Disambiguation5/5

    Each tool has a clearly distinct resource/action pairing: jobs, pods, job configs, pod configs, and two log types. Pipeline worker logs and service logs are the only overlapping pair, but the names and descriptions cleanly separate them.

    Naming Consistency5/5

    All tools follow a consistent verb_noun pattern using get_ or list_ prefixes. The resource nouns are uniform and predictable, with job/config/pod/log/service modifiers making the hierarchy easy to infer.

    Tool Count5/5

    Eight tools is a well-scoped size for a pipeline inspection server. Each tool covers a distinct need without redundancy or surface bloat.

    Completeness4/5

    The set provides solid read-only coverage of jobs, pods, configs, and logs, which matches an observability-focused purose. Minor gaps exist, such as no direct way to list services or aggregate request-level status, but agents can work around these using the existing list and get tools.

    Maintenance

    ActivityStale
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables read-only Kubernetes incident investigation through MCP tools for listing pods, describing resources, fetching logs, and searching runbooks.
      1
      -
    • A
      license
      B
      quality
      C
      maintenance
      Enables users to inspect ephemeral GitHub Actions runner fleet status, SLOs, incidents, capacity recommendations, and cost evidence through read-only MCP tools, without allowing infrastructure or cluster mutations.
      5
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables read-only access to Autobase Console projects, clusters, environments, operations, PostgreSQL versions, and extensions through configurable MCP tools. Supports paginated listing and filtering, with limited redacted operation logs and no write or arbitrary execution tools.
      MIT