multivon-mcp

Docs · Website · PyPI · multivon-eval (engine)

These 22 tools are what an autonomous eval agent needs to do its job: discover its own capabilities (eval.discover), normalize traces from any source (ingest_trace), and run calibrated evaluators against them. The framework lives behind an MCP boundary because that's the future shape of eval — a swarm of specialized eval agents coordinating through the protocol, not a SaaS dashboard.

MCP server that gives AI coding agents direct access to evaluation tools. Drop into Claude Desktop, Claude Code, Cursor, Cline, or any Model Context Protocol–compatible agent.

When the agent is helping you build an LLM product, it can:

• Score a RAG output for hallucination without you writing the scaffolding

• Generate an adversarial PDF on demand to test your document AI

• Run the full pdfhell mini-suite against a model and analyse the results

• Produce a hash-chained audit pack for procurement diligence

• Discover the full evaluation capability catalog as JSON

No copy-paste, no python -c "...", no asking the agent to figure out the SDK calls.

Install

pip install multivon-mcp

Bare install pulls multivon-eval, pdfhell, and the MCP SDK. The provider SDKs (anthropic, openai, google-genai) come along too — bring your own API key in env.

Configure your agent

Claude Desktop / Claude Code

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "multivon": {
      "command": "multivon-mcp",
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-...",
        "OPENAI_API_KEY": "sk-proj-...",
        "GOOGLE_API_KEY": "AIza..."
      }
    }
  }
}

Restart Claude. The 22 tools become available; ask Claude "use multivon to evaluate this RAG output" and it figures out which tool to call.

Cursor

cursor.json or via Settings → MCP:

{ "mcpServers": { "multivon": { "command": "multivon-mcp" } } }

Cline / OpenCode / any MCP-compatible agent

Same shape — point at the multivon-mcp console script.

Local dev / debugging

mcp dev multivon_mcp.server

Opens the MCP Inspector UI in your browser. You can call any tool by name, see the JSON schemas, and watch the requests/responses.

The 22 tools

Discovery & document AI

RAG generation & retrieval

Safety, compliance, fairness

Agent & multimodal

Agent traces. eval_tool_call_accuracy and the other agent-trace evaluators in multivon-eval (ToolArgumentAccuracy, ToolCallNecessity, TrajectoryEfficiency, AgentMemoryEval, PlanQuality, TaskCompletion, StepFaithfulness) take an agent_trace=[AgentStep(...)] plus expected_tool_calls=[...] on the case. Three-shape semantics matter: expected_tool_calls=None skips, [] asserts "no tools called", and [...] checks the trace contains the named calls in order. The MCP tool wraps this — pass the trace JSON via eval_ingest_trace first to normalize it from LangGraph / OpenAI Agents SDK / manual shapes. See the multivon-eval agent integrations for the source-of-truth tracer code.

Flexible scoring

Agent workflows (new in 0.3.0)

Example session

User: I just shipped a RAG endpoint. Can you check it for hallucinations?

Claude: I'll use multivon to evaluate it.
        [calls eval_discover to see what's available]
        [calls eval_faithfulness with your input/context/output]

→ score: 0.667 (passed: False), threshold: 0.9
  reason: 2/3 claims grounded
    ✓ "annual renewal" — supported by context
    ✓ "30-day notice" — supported by context
    ✗ "automatic upgrade" — NOT in context

Claude: Your RAG hallucinated the "automatic upgrade" detail. The context
        doesn't mention upgrades. I'd add a Hallucination evaluator to your CI
        gate, threshold ≥0.85, and re-prompt with explicit "only use facts
        from context" instructions.

Why these 22 tools (not all 44)

eval_discover returns the full 44-evaluator catalog, so the agent can always introspect everything. The 22 tools we expose directly are the ones agents actually call mid-edit:

• RAG generation checks (faithfulness, hallucination, relevance, answer_accuracy)

• RAG retrieval checks (context_precision, context_recall)

• Safety / fairness guardrails (toxicity, bias)

• Compliance (pii_detection, schema_compliance) — local-only, no API egress

• Flexible scoring (g_eval, custom_rubric) for user-defined rubrics

• Multimodal (vqa_faithfulness, document_grounding) for vision agents

• Agent traces (tool_call_accuracy)

• Document AI (pdfhell.run, pdfhell.make) — for any RAG-on-PDFs flow

• Audit pack — when procurement is involved

• Discover — meta-capability for planning

• Agent workflows (compare_runs, generate_cases, ingest_trace) — the loop that turns one-shot scoring into iterative improvement

The three new 0.3.0 tools matter because evals are most useful as a loop, not a single call: generate a starting suite from your own docs (eval_generate_cases), run your agent over it, score the trace (eval_ingest_trace → eval_*), make a fix, then verify the fix improved things vs. the baseline (eval_compare_runs). Agents need that whole loop callable from within a conversation — otherwise they fall back to ad-hoc judgment.

Exposing all 44 evaluators as MCP tools would bloat the agent's context window and overwhelm tool-selection. If you need an evaluator that's not directly exposed, the agent can still use multivon-eval as a library — eval_discover returns the import paths.

Dependencies

Hard pins (from pyproject.toml):

• mcp[cli] >= 1.0 — official MCP Python SDK + the mcp dev inspector

• multivon-eval >= 0.9.4 — the evaluator surface this wraps

• pdfhell >= 0.1.0 — the adversarial-PDF benchmark this wraps

Recommended (effective floor for full feature parity):

• multivon-eval >= 0.9.8 — pulls in the corrected calibrated-threshold logic from the 0.9.7 hotfix (which affects what eval_discover reports and any tool that surfaces benchmark numbers in its docstring), plus the bundled Claude Code skills + multivon-eval install-skills CLI from 0.9.8.

• pdfhell >= 0.5.4 — pulls in the mini-v4 17-trap suite and the pdfhell.research autoresearch loop. The pdfhell_run --suite mini-v4 tool path assumes these are present.

The pyproject pins are kept loose so existing deployments don't break; pin the recommended floors yourself if you care about the corrected benchmark numbers or the new suites.

All Apache 2.0.

MCP server vs Claude Code skills vs eval-action — which one do I use?

multivon-eval ships three agent-facing surfaces. They overlap on what they call (the same evaluator catalog) but differ on where the agent lives.

If you're building an LLM product and want the agent in your editor to score a RAG output without copy-pasting Python, use multivon-mcp. If you live in Claude Code and want the bootstrap → audit → explain loop wired up as native commands, use the bundled skills. If you want PR-time gating, use the GitHub Action. The three are complementary — most projects end up using all three.

The Multivon ecosystem

Five public + one early-access package, all built on a shared evaluation engine:

License

Apache 2.0.

Citing

@software{multivon_mcp,
  title  = {multivon-mcp: MCP server exposing multivon-eval + pdfhell as agent-callable tools},
  author = {Multivon},
  year   = {2026},
  url    = {https://github.com/multivon-ai/multivon-mcp},
}