Weavatrix — native MCP repository intelligence

Give your coding agent repository evidence before it starts guessing.

Weavatrix answers the questions agents otherwise invent: what breaks if this file changes, who consumes this n8n field, whether an MCP schema still accepts yesterday’s request, which viem call sites follow an ABI layout change, and whether .weavatrix/architecture.json still holds. 67 read-only operations, one revision-bound graph, no LSP and no prompt pack.

Codex, Claude Code, Cursor, and Grok get impact, architecture, APIs, Git history, duplicates, dead code, search, semantic links, temporal memory, exported n8n workflows (n8n_*), Dify YAML apps (dify_*), Agent Plugins/Skills/MCP catalogs (agent_*), Mermaid flowcharts (diagram_*), and Web3 ABI/consumer impact (web3_*). Those parsers are domains on Weavatrix Core, not new MCP products. See n8n, Dify, and parsers.

It does not answer from a larger grep or an invented confidence score. Every bounded result can carry the repository revision, file, line, extractor, evidence kind, and confidence that produced it.

This npm package is the convenient prebuilt native tarball. That is the only reason it exists on npm. It is not a TypeScript rewrite, not a Serena wrapper, and not a Repomix-style packer with an MCP flag. The crate weavatrix on crates.io is the same adapter; weavatrix-js is the legacy JavaScript implementation used for historical baselines and is not bundled here.

Install this package when the machine should run MCP without a Rust toolchain. Link weavatrix-rust when you are embedding the engine. Use the Cursor/Codex/Claude plugin when the workspace folder must be the root, not npx’s current directory.

Install in 30 seconds

npx -y weavatrix mcp .

Or install the same native MCP product through Cargo:

cargo install weavatrix
weavatrix mcp .

Codex

[mcp_servers.weavatrix]
command = "npx"
args = ["-y", "[email protected]", "mcp", "."]

Pass an absolute repository path when the Codex process cwd is not the project you intend to analyze.

Claude Code

claude mcp add weavatrix -- npx -y [email protected] mcp .

Cursor

Prefer the plugin, or pin the open workspace explicitly and do not also enable the plugin for the same server name:

{
  "mcpServers": {
    "weavatrix": {
      "command": "npx",
      "args": ["-y", "[email protected]", "mcp", "${workspaceFolder}"]
    }
  }
}

Profiles expose bounded views of the same engine:

npx -y [email protected] mcp . --profile=all
npx -y [email protected] mcp . --profile=code
npx -y [email protected] mcp . --profile=seo
npx -y [email protected] mcp . --profile=n8n
npx -y [email protected] mcp . --profile=dify
npx -y [email protected] mcp . --profile=agent
npx -y [email protected] mcp . --profile=diagram
npx -y [email protected] mcp . --profile=web3

The package contains native binaries for Windows x64/arm64, macOS x64/arm64, and glibc Linux x64/arm64. It has no install script and performs no runtime download.

Halve every answer on a client that reads structured output

An MCP result carries the payload twice: once as structuredContent, and once mirrored into a text block for clients that read only content. The mirror is the pretty-printed copy, so it is the larger of the two.

npx -y weavatrix mcp . --output-format=structured

Measured on run_audit over a real repository, the response falls from 8931 to 3589 bytes, 59.8% smaller. Whether a client reads structured output does not change between calls, so it is chosen once at startup rather than restated as an argument on every call; WEAVATRIX_OUTPUT_FORMAT=structured does the same, and a call that names its own output_format still wins.

json is the default and keeps the mirror, because a client that ignores structuredContent would otherwise see an empty result. For an architecture question, call architecture_inventory with no arguments and explain its modular, onion, layered, and ports-and-adapters hypotheses in prose. Report CANDIDATE and INSUFFICIENT_EVIDENCE honestly; the declared contract style is not evidence of the observed architecture.

What the new domains actually answer

| Ask | Tool | Honest limit | | --- | --- | --- | | Will this MCP schema still accept yesterday’s request? | agent_change_impact | string → integer is incompatible. A missing tool in a partial catalog is unconfirmed, not removed. | | Who calls this ABI after an event layout change? | web3_impact | Comment/string “calls” are not consumers. ABI equality is not a live deployment. | | Who reads this field in an exported n8n workflow? | n8n_trace | Secrets stay off the graph. | | Which Dify nodes consume start_node.query? | dify_trace | Conversation variables are directed edges, not string presence. | | Does this Mermaid arrow prove a code call? | diagram_* | Arrows are declared_architecture, never Calls. | | Does .weavatrix/architecture.json still hold? | verify_architecture | Unknown rules fail closed. |

What an agent can ask

What breaks if I change src/auth/middleware.ts?
Trace POST /api/orders through this backend and its clients.
Which production symbols are dead, and what evidence proves it?
Show duplicate implementations but suppress router boilerplate.
Which dependency violates .weavatrix/architecture.json?
Find every GraphQL, gRPC, Kafka, RabbitMQ, NATS, JMS, SQS, or SNS
contract affected by this branch.
Build the smallest source bundle needed to edit this symbol safely.
Show me this file as it was two commits ago, without a checkout.
If I change the email recipient in this exported n8n workflow, who reads it?
If I change start_node.query in this Dify YAML, which nodes consume it?
Did this MCP catalog make `id` an integer so old string requests fail?
Which viem `decodeEventLog` sites still assume the old Deposit layout?
Is this Mermaid arrow a real call, or only declared architecture?

See it answer

Real answers from Weavatrix analyzing its own repository at commit ec8bf30, abridged (…) and with the local root shortened. Reproduce any of them with weavatrix tool <name> . '<arguments>'.

Orient in one call — and know exactly which repository and revision answered:

{"name": "graph_stats", "arguments": {}}

{
  "nodes": 1294,
  "edges": 2771,
  "freshness": {"state": "CURRENT", "source_revision": "sha256:dd3d96d3…"},
  "relations": {"calls": 651, "contains": 1271, "imports": 419, "references": 314, "…": "…"},
  "repository_context": {
    "root": "…/weavatrix",
    "git_head": "ec8bf3041588623f23ec5e7ebbaff3a333da9ca7",
    "scan_revision": "sha256:dd3d96d3…",
    "graph_age_seconds": 0
  }
}

Every answer carries that repository_context block. From 1.11.1, each MCP process pins its launch root for the session: open_repo to another path is refused unless you started with --allow-retarget, and omitted expected_repository is filled from that pin. Still pass "expected_repository": "weavatrix" when you already know the folder name — a mismatched server fails loudly instead of answering about the wrong repository.

Blast radius before you edit:

{"name": "get_dependents", "arguments": {"label": "file:src/mcp/server/mod.rs", "max_nodes": 6}}

{
  "dependents": [
    {"distance": 1, "node": {"id": "file:src/main.rs", "language": "rust"}},
    {"distance": 1, "node": {"id": "file:src/mcp/mod.rs"}},
    {"distance": 1, "node": {"id": "file:src/mcp/server/tests/catalog.rs"}},
    {"…": "…"}
  ]
}

The file as it was — no checkout (git_read_blob):

{"name": "git_read_blob", "arguments": {"path": "Cargo.toml", "revision": "HEAD~3", "max_bytes": 400}}

{
  "path": "Cargo.toml",
  "revision": "70b5bc788a10ec89ab28b9ecbc005e3b9c7f9829",
  "oid": "f5f54bac00a4feb6e965c0d87c5c6d4d25782e91",
  "kind": "utf8-text",
  "lines": ["[package]", "name = \"weavatrix\"", "version = \"1.9.2\"", "…"],
  "size_bytes": 1224,
  "returned_bytes": 400,
  "truncated": true
}

Three commits before ec8bf30 this package was 1.9.2; the agent reads that follow-up to a diff without touching the worktree. Binary blobs fail closed instead of being decoded into garbage.

The 67 read-only operations

| Workflow | Operations | | --- | --- | | Graph orientation | graph_stats, get_node, get_neighbors, query_graph, god_nodes, shortest_path, get_community, list_communities, module_map, build_graph | | Change impact | get_dependents, change_impact, select_tests, verified_change, prepare_change, graph_diff | | Exact source context | search_code, read_source, inspect_symbol, go_to_definition, find_references, context_bundle, map_stacktrace | | Health and quality | find_duplicates, find_dead_code, run_audit, coverage_map, hot_path_review, perf_attribution | | APIs and transports | list_endpoints, trace_endpoint, trace_api_contract | | Architecture | architecture_inventory, get_architecture_contract, verify_architecture, verify_capabilities, explain_architecture_violation, propose_architecture_exception | | Local CI | ci_restrictions, explain_restriction | | Git and repositories | git_history, git_read_blob, cross_repo_git, open_repo, list_known_repos, rebuild_graph | | Native extensions | vector_search, semantic_link, seo_link_suggestions, memory_context | | n8n workflows | n8n_inventory, n8n_trace, n8n_context | | Dify apps | dify_inventory, dify_trace, dify_context | | Agent packages | agent_inventory, agent_trace, agent_context, agent_change_impact | | Mermaid diagrams | diagram_inventory, diagram_trace, diagram_context | | Web3 integration | web3_inventory, web3_trace, web3_impact, web3_context |

Every operation is read-only with respect to the analyzed repository. Pagination and explicit limits bound large neighborhoods, histories, searches, and contract inventories.

build_graph and architecture_inventory share one typed build model. They also report independent, evidence-bounded hypotheses for modular source organization, onion, ports-and-adapters, layered dependencies, and deployment uncertainty; the declared target style is never observation. The summary includes each hypothesis's status, supporting signals, contradictions, and unknowns; use detail:"full" for paged graph evidence. They recognize Rust impl Trait for Type as an explicit implementation signal, without mistaking inherent impl Type for one. They preserve nested Cargo, npm/TypeScript, Go, and Python components, many-to-many declared memberships, collision-free identities, configuration predicates, and immutable source evidence. Totals and SCC/cycle witnesses are computed before pagination, so a response cap cannot change the graph verdict.

ci_restrictions reuses that target-aware topology while reading local GitHub Actions workflows and literal checks. Root files, repeated commands, UTF-8 BOM offsets, optional events, base branches, and changed-path scenarios keep exact evidence. Dynamic conditions, mutually exclusive configuration unions, actual runs, and remote required checks remain explicit unknowns; a workflow file alone never establishes a passing gate.

Need measured coverage? Run Weavatrix Quality first, then coverage_map. The host will not invent percentages from an empty search path. Quality publishes .weavatrix/coverage/lcov.info. Agent schema verdicts are a supported-subset proof: unsupported keywords stay undetermined, and skill hits are declared allowed-tools. Web3 consumers stay on the paired files; a missing candidate is missing_input.

24 repository surfaces

| Group | Surfaces | | --- | --- | | Code | Rust; JavaScript/JSX; TypeScript/TSX; Python; Go; Java; C#; C; C++; SQL; Bash/Zsh; Swift; Solidity | | Contracts and configuration | GraphQL; Protobuf/gRPC; JSON/JSONC (n8n after parse); YAML/Kubernetes (Dify after parse); Terraform/HCL; XML | | Documents and UI | HTML/Vue/Svelte; CSS/SCSS/Sass/Less; Markdown/MDX; reStructuredText; AsciiDoc |

Cross-surface analysis connects HTTP, GraphQL, gRPC, Kafka, RabbitMQ/AMQP, JMS, NATS, SQS, and SNS evidence. Dynamic dispatch that cannot be proved stays unresolved; static reachability is never presented as measured coverage.

Workflow export domains

n8n and Dify are after-parse domains on the same graph, not new MCP servers.

export in git → existing JSON/YAML parse → typed domain → inventory / trace / context

| Domain | Tools | Status | | --- | --- | --- | | n8n | n8n_inventory, n8n_trace, n8n_context | Product 1.12.0 | | Dify | dify_inventory, dify_trace, dify_context | Product 1.13.0 | | Agent packages | agent_inventory, agent_trace, agent_context, agent_change_impact | Product 1.14.0 | | Mermaid diagrams | diagram_inventory, diagram_trace, diagram_context | Product 1.14.0 | | Web3 integration | web3_inventory, web3_trace, web3_impact, web3_context | Product 1.16.4 |

| Need | Use | | --- | --- | | Change impact on an export in git | Weavatrix n8n_* / dify_* | | Live n8n instance | n8n-mcp | | Live Dify console | dify-mcp | | CI lint/diff of Dify YAML | difyctl |

No invented n8n/Dify millisecond benches. Those tools share the measured installed MCP boundary (2026-08-03): cold 32.21x (157.34 ms vs 5,068.22 ms) and warm 36.85x (7.94 ms vs 292.55 ms) versus weavatrix-js 0.3.15.

Product and engine are separate

coding agent
    |
    | MCP over stdio
    v
weavatrix 1.17.5
    profile catalog · refresh · watcher · MCP framing
    |
    v
weavatrix-rust 2.17.4
    typed graph · analysis · 67 product operations including local CI, n8n, Dify, agent packages, Mermaid, and Web3

This npm product owns MCP transport and native distribution. The weavatrix-rust crate is the reusable protocol-independent engine; it is not an MCP server. weavatrix-stream provides adapters for Weavatrix observations; this host does not take a runtime dependency on it. Its standalone diagnostic therefore reports weavatrix-rust <engine-version>, while this MCP product reports both its product and embedded-engine identities.

You do not need the server to use the engines

The layers underneath are separately released npm packages with their own repositories, licenses, and TypeScript types. Each one is a native library you can require directly, with no MCP server, no agent, and no dependency on any other Weavatrix package:

| Package | What it does on its own | | --- | --- | | weavatrix-scan | Deterministic repository manifests: normalized paths, sizes, hashes, one revision, typed skip evidence. | | weavatrix-parse | Lossless tokenization and structural facts for 25 languages, with exact spans. | | weavatrix-graph | A directed graph with provenance on every edge, plus BFS, SCC, topological sort, and PageRank. | | weavatrix-memory | Bitemporal, evidence-carrying agent memory with a hard token budget and an audit receipt. | | weavatrix-search | Bounded ignore-aware content search, plus a persistent index updated from watcher events. | | weavatrix-search-vector | Persistent, mutable vector search with metadata filters and an exact oracle. | | weavatrix-clone | Type-1/2/3 clone detection with per-pair evidence and JSON, SARIF, and BigCloneEval output. |

Their Node and Bun benchmarks against the JavaScript library each would otherwise replace live in one place, with the rules written down and the losing rows kept: weavatrix-benchmarks.

Release evidence

The installed-package benchmark packs both products, installs them into isolated npm roots, starts fresh MCP processes with empty caches, and validates identity, advertised operations, results, and cleanup.

The packaged 1.2.0 product (weavatrix-rust 2.1.1) was measured on 2026-08-03 against installed weavatrix-js 0.3.15 on a real JavaScript service repository: paired cold-boundary median 32.06x (spawn to first tool result: 157.34 ms vs 5,068.22 ms) and warm tools/call median 36.85x (7.94 ms vs 292.55 ms), passing the 24x cold and 30x warm release thresholds and sitting slightly above the 30.34x recorded for the 1.0.0 baseline.

Full evidence and methodology: benchmarks.

Safety

• read-only MCP surface;
• no repository-code execution or source writes;
• no network path in analysis;
• no npm install script or runtime binary download;
• bounded inputs, outputs, histories, and pagination;
• stable ordering and revision provenance;
• unsafe Rust forbidden in first-party engine crates;
• MIT licensed.

Links

• Source and issues
• Engine API
• Tool reference
• Language support
• Architecture
• n8n · Dify · parsers
• Benchmarks

License

MIT.