Agenda Intelligence Md

What vassiliylakhonin/agenda-intelligence-md MCP server does

The vassiliylakhonin/agenda-intelligence-md MCP server provides deterministic checks for claim-backed AI outputs and structured strategic-risk workflows. It accepts caller-supplied JSON, source text, evidence records, or agenda requests, then returns validation results, evidence gaps, support classifications, review actions, or a structured analysis result.

Its scope includes sanctions, regulatory, geopolitical, trade, corridor, policy, maritime, market-entry, and agent-interaction risk. The package also exposes static protocol instructions, schemas, regional and sector lenses, source-category requirements, and archived signal records. These resources can help an agent construct requests and apply a consistent evidence process without depending on a live search service.

The vassiliylakhonin/agenda-intelligence-md MCP server is appropriate for local CI checks, agent review loops, memo preparation, and pre-action evidence triage. Results remain subject to human review, particularly for legal, sanctions, compliance, financial, investment, authorization, or other high-consequence decisions.

How it works

Most tools operate only on information supplied by the caller. Schema validators check whether briefs, evidence packs, audits, requests, or memos conform to packaged JSON schemas. Evidence-oriented tools compare claims with declared source IDs, supplied source text, quotations, numbers, and lexical patterns. They can identify missing references, quote mismatches, weak textual support, polarity conflicts, unmatched numbers, orphan evidence, and unsupported claims.

Source planning and coverage tools use packaged source-category requirements. They report which source types are required, matched, or missing; they do not find or fetch those sources. Claim verdicts can assess caller-provided evidence against declared freshness, authority class, independent source groups, conflicts, jurisdiction, and subject identifiers, but a verified result means only that the configured evidence threshold was met.

The analyze tool generates an auditable strategic-risk memo from a structured Agenda request. If ANTHROPIC_API_KEY is unset, it returns the assembled system prompt for the host model to complete rather than completing the analysis itself. Vertical tools return bounded triage outputs such as decision-readiness scores, evidence gaps, risk signals, owner actions, and watch-next indicators.

Setup and configuration

The project is distributed on PyPI as agenda-intelligence-md and is licensed under MIT. The README documents installation from a source checkout with an editable Python install, and installation of the pinned release with pip. It also documents a CLI named agenda-intelligence for checking evidence packets, discovering possible source matches, and reviewing local Markdown, DOCX, or PDF files. PDF extraction requires the optional documents extra.

The repository provides MCP-facing functionality alongside CLI, Python API, A2A, and serverless-oriented examples. The supplied material does not specify a single MCP client configuration or a dedicated MCP launch command. No credential is required for local evidence checks; an Anthropic API key may affect how analyze behaves when a host model is expected to finish the returned prompt.

Tools and capabilities

• Validate agenda briefs, evidence packs, audits, and memos against packaged schemas.
• Check evidence packets, quoted fragments, claim grounding, source coverage, and output relay safety.
• Load protocols, schemas, regional or sector lenses, source categories, and archived signals.
• Plan evidence requirements and append claims incrementally to an evidence pack.
• Generate strategic-risk memos and run bounded vertical triage for corridor, sanctions, maritime, market-entry, and agent-trust scenarios.
• Route proposed actions to continue, request evidence, require approval, or stop using caller-supplied inputs.

Limitations and notes

The vassiliylakhonin/agenda-intelligence-md MCP server does not autonomously retrieve current sources, browse the web, discover documents, or update its archived signals. Lexical grounding and quote checks do not prove factual accuracy, source reliability, semantic equivalence, reversed roles, or identity. The tools do not provide legal, sanctions, compliance, financial, investment, cybersecurity, authorization, or launch-approval decisions. Human review remains required, and deep_dive is a future placeholder rather than a production analysis workflow.