PMAgent-Spec MCP Server

PMAgent-Spec is an MCP (Model Context Protocol) server that delivers PMAgent content specs (Monday Minutes, revision histories, OKR reports, SDK READMEs, etc.) to GitHub Copilot or any MCP-capable client. The companion VS Code extension auto-registers the hosted server and drops the pmagent-orchestrator agent template into your repo so Copilot can immediately call the PMAgent tools.

Install in VS Code

1. Open the Extensions view (Cmd/Ctrl+Shift+X).
2. Search for “PMAgent Spec MCP” and install.
3. The extension registers https://pmagent-spec.nicesmoke-219d4124.eastus2.azurecontainerapps.io/mcp automatically. Check the PMAgent MCP output channel for confirmation.

MCP helpers to pair with

• GitHub MCP: Use when repos and PRs live in GitHub (merged PRs, blockers, milestones, upcoming project items). Repo
• Azure DevOps MCP: Use when telemetry lives in ADO (Boards, pipelines/releases, blockers, upcoming backlog items). Repo
• Power BI Remote MCP: The remote Power BI MCP server enables AI agents to query Power BI semantic models using natural language. (e.g., OKR or Monday Minutes KPIs). Repo
• Kusto/ADX MCP: Use Azure MCP

Use with GitHub Copilot in VS Code

1. In Copilot Chat, choose the pmagent-orchestrator agent (the extension adds .github/agents/pmagent.agent.md for you).
2. Describe the deliverable plus telemetry sources (GitHub vs ADO vs Power BI) and the repos/projects/date range.

Supported scenarios (specs)

• monday_minutes: Generate Monday Minutes updates for a team or project.
• revision_history: Create revision history / release notes from GitHub PRs and Azure DevOps work items tied to a release.
• product_status_report: Summarize product/release health with progress, risks, bugs, dependencies, and next steps.
• feature_history_traker: Turn commits/PRs/work items into a narrative feature history grouped by phases/workstreams.
• sdk_readme: Generate or update an SDK README from source code, APIs, dependencies, and samples.
• powerbi_telemetry_knowledge: Generate knowledge of PowerBI Semantic Model Schema and product-specific OKRs.
• adx_telemetry_knowledge: Generate knowledge from ADX (Azure Data Explorer) dashboard json file including predefined queries and filters.
• okr_report: Produce a report with OKR and corresponding analysis/insights from PowerBI semantic model or ADX KQL queries.
• spec_creation: Collect user inputs per templates/spec_template.md and generate a new spec that matches that structure.
• Template injection: Specs can embed {{template_name.md}} placeholders; fetch_spec will inline the matching file under templates/ (local first, remote fallback) before returning.

OKR Report

• Chat with pmagent-orchestrator agent in Copilot to build your product-specific knowledge:
  • PowerBI Semantic Model Schema and OKRs Knowledge: Copilot will use Power BI Modeling MCP and powerbi_telemetry_knowledge spec from PM Agent to help you generate the initial knowledge draft.
  • ADX Kusto Query Knowledge: Copilot will use Kusto tools from Azure MCP and adx_telemetry_knowledge spec from PM Agent to help you generate the initial knowledge draft.

create a new PowerBI telemetry knowledge markdown file for <your-product-full-name> based on <links-to-your-product-dashboard>

• PM Agent cannot doscover OKRs from the dashboard directly. Therefore, you may also need to revise the product-specific knowledge in these areas:

  • IsAnalysisDimension for each column: If a column is tagged Yes in IsAnalysisDimension, the column should be treated as a primary candidate for cohort analysis and can be safely used to break results down into slices (segments) for comparisons across groups.
  • Entity Relationships
  • Definiation of each OKR(s), including Time Semantics, Telemetry Scope and Filters, Metric Calculation Logic, Targets, Thresholds, and Guardrails and Recommended Breakdowns and Dimensions.

• Chat with pmagent-orchestrator agent in Copilot to generate a new OKR report for your product. Copilot will use Power BI Modeling MCP or Azure MCP (with Kusto) and okr_report spec from PM Agent to help retrive real data from PowerBI dashboard or Kusto queries and generate OKR report.

generate new okr report for <your-product-name>> based on <your-product-specific-knwoledge-on-local> and write into a new markdown file

Remote MCP config (manual)

If you are not using the extension, register the hosted server in your MCP client (e.g., ~/.config/github-copilot/mcp.json):

{
  "mcpServers": {
    "pmagent-spec": {
      "transport": {
        "type": "http",
        "url": "https://pmagent-spec.nicesmoke-219d4124.eastus2.azurecontainerapps.io/mcp"
      }
    }
  }
}

Add your other MCP servers (GitHub, Azure DevOps, Power BI) to the same file per their docs so the manifests in tool_specs/ can be used.

What it does

• Serves spec templates and writing guardrails through content_generation_best_practice, list_specs, and fetch_spec.
• Documents telemetry helpers through get_tool_manifest, mapping each spec to the right GitHub, Azure DevOps, or Power BI MCP calls.
• Provides an extension that registers the remote server at activation (no manual JSON edits required).

Local development

1. Optional: create a virtual environment in src/ (python -m venv .venv && source .venv/bin/activate).
2. Install dependencies: pip install -r src/requirements.txt.
3. Start the MCP server: python src/server.py (listens on http://0.0.0.0:8100/mcp).
4. Point your MCP config or the VS Code setting pmagentSpecMcp.serverUrl to http://localhost:8100/mcp, then reload the client.

Contributing

See CONTRIBUTING.md.