Skip to main content
Glama

etch-mcp

etch-mcp MCP server Listed on mcpservers.org

Etch is a signed audit chain for AI agent decisions, offline-verifiable against pinned public keys.

This repository is the MCP integration surface for Etch, the hosted service. It carries only:

  • The MCP registry metadata (server.json, glama.json)

  • Per-client integration examples for Claude Code / Cursor / Continue / Cline / Codex

  • Documentation for attaching an MCP-compliant client to Etch

The hosted Etch service source is proprietary and lives in a separate repository. Nothing in this repository is required to be inspected or modified in order to use Etch.

Try in 30 seconds, no signup

curl -X POST https://etch.systems/v1/your-project

The response returns a bearer token, an MCP endpoint URL, and an anonymous project id. Point any MCP-compliant client at https://etch.systems/mcp with the token. Anonymous projects carry 500 signed events over 14 days; convert to a persistent project any time to keep the chain forever.

Full walk-through: etch.systems/docs/quickstart.

Related MCP server: DCL Evaluator

Per-client config

Claude Code

~/.claude/settings.json:

{
  "mcpServers": {
    "etch": {
      "url": "https://etch.systems/mcp",
      "headers": {
        "Authorization": "Bearer wm_YOUR_ANON_TOKEN"
      }
    }
  }
}

Cursor

~/.cursor/mcp.json (or .cursor/mcp.json at the project root):

{
  "mcpServers": {
    "etch": {
      "url": "https://etch.systems/mcp",
      "headers": {
        "Authorization": "Bearer wm_YOUR_ANON_TOKEN"
      }
    }
  }
}

Continue

~/.continue/config.yaml:

mcpServers:
  - name: etch
    url: https://etch.systems/mcp
    headers:
      Authorization: Bearer wm_YOUR_ANON_TOKEN

Cline (VS Code)

~/.config/Cline/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json:

{
  "mcpServers": {
    "etch": {
      "url": "https://etch.systems/mcp",
      "headers": {
        "Authorization": "Bearer wm_YOUR_ANON_TOKEN"
      }
    }
  }
}

Codex CLI

~/.codex/config.toml:

[mcp_servers.etch]
url = "https://etch.systems/mcp"
headers = { Authorization = "Bearer wm_YOUR_ANON_TOKEN" }

Per-client documentation with current config file locations is kept up-to-date at etch.systems/docs/mcp-clients.

Verify

Every event Etch signs is verifiable offline against pinned public keys with the OSS reference verifier. Auditors pin fingerprints once at pilot onboarding; every past epoch is verifiable indefinitely without contacting Etch.

What Etch is NOT

  • Not a media-provenance envelope (that lane is C2PA and SPIF). Etch signs decisions, not content. Comparison: etch.systems/vs/c2pa.

  • Not a transparency log (that lane is Sigstore Rekor). Etch is a per-project chain that anchors epochs on Rekor. Comparison: etch.systems/vs/sigstore-rekor.

  • Not a watermarking system. Etch records who reviewed an output and what they applied, not whether a specific model touched the text.

License

The metadata and documentation in this repository are MIT-licensed (see LICENSE). The hosted Etch service is Business Source License 1.1 with a hosted-service restriction; see etch.systems/pricing for terms.

Registry listings

Etch is listed on:

PulseMCP auto-indexes from the Official MCP Registry when their intake resumes; no separate submission needed.

For issues with the hosted service, contact via etch.systems. For issues with the metadata or integration examples in this repository, open a GitHub issue here.

Available Tools

31 tools
export_claude_mdBInspect

Generate a CLAUDE.md document from the knowledge graph (top constraints, recent decisions, known bug regions, co-edit patterns).

Inputs: max_constraints.

find_contradictionsCInspect

Find pairs of facts that contradict each other based on similarity and status differences

Inputs: limit, query.

get_agents_md_constraintsAInspect

Parse AGENTS.md / CLAUDE.md / GEMINI.md / .agents/skills/*.md in the project and return declarative constraints. Mixed into PreToolUse enforcement automatically; this tool exposes the same data for inspection.

Inputs: file_path, project_dir.

get_audit_log_headAInspect

v0.13 tamper-evident audit log. Return the current head state (last log entry seq, last closed epoch seq, unclosed-entry count) plus the full closed-epoch chain with hybrid signature envelopes. Compliance auditors call this periodically to verify no operator misbehavior has occurred since the last check. Requires WORLD_MODEL_AUDIT_LOG=on at server startup.

Inputs: no named parameters.

get_co_edit_suggestionsBInspect

Get files commonly edited alongside the given file based on historical patterns

Inputs: file_path (required), limit.

get_compaction_auditAInspect

List recent compaction audit entries, most-recent first. Filter by session_id or limit count.

Inputs: limit, session_id.

get_constraintsAInspect

Get constraints (linting rules, patterns, conventions) for a file

Inputs: file_path (required), constraint_types.

get_context_for_actionCInspect

Pre-action context bundle: constraints, decisions, bugs, co-edits, related facts, and risk score for a file before editing

Inputs: file_path (required), action_type (required).

get_decision_logCInspect

Get decision traces showing agent proposals and human corrections

Inputs: limit, file_path, session_id, decision_type.

get_health_reportAInspect

Memory health diagnostics: orphans, stale facts, contradictions, decay candidates, DB sizes

Inputs: no named parameters.

get_injection_contextBInspect

Return a compact constraint+fact bundle for PostCompact / UserPromptSubmit hooks to re-inject after context loss.

Inputs: event_type (required), max_facts, project_hint, max_constraints.

ingest_pr_reviewsAInspect

Pull GitHub PR review comments and convert them into learned constraints in the knowledge graph

Inputs: repo, count.

pin_annotationAInspect

Attach a signed human annotation (note, override rationale, or intervention record) to a span of agent events. Persists into the annotations table and chains into the same Merkle audit log as agent writes (v0.15.0, ADR-0001). Rationale limited to 8 KB.

Inputs: author (required), rationale (required), session_id (required), annotation_type (required), +2 more.

predict_regressionAInspect

Score regression risk for a proposed change to a file based on past bugs, test failures, and constraint violations

Inputs: file_path (required), change_description.

predict_test_failuresAInspect

Surface tests likely to fail given a set of edited files

Inputs: file_paths (required).

promote_constraintBInspect

Promote a constraint from this project to all other registered projects

Inputs: constraint_id (required), target_projects.

prove_entry_inclusionAInspect

v0.13 tamper-evident audit log. Return a cryptographic inclusion-proof bundle for a persisted row_id (fact, constraint, event, or decision ID). Bundle includes the entry, the containing signed epoch (Ed25519 + SLH-DSA hybrid signature envelope), an RFC 6962 Merkle inclusion proof, and the full epoch chain from genesis. Requires WORLD_MODEL_AUDIT_LOG=on at server startup; returns an error object when opt-in is off, when the row_id is not found, or when the entry is in the unclosed backlog.

Inputs: row_id (required).

query_factBInspect

Query the knowledge graph for facts about entities (APIs, functions, classes, etc.)

Inputs: query (required), context, entity_type, content_type.

recall_transcript_rangeCInspect

Hydrate a Claude Code session transcript by line range. Lets agents trace a fact back to the exact conversation that produced it.

Inputs: session_id (required), line_end, line_start.

record_compaction_auditAInspect

Record a context-compaction event with token counts and what was re-injected. Lets developers audit what was remembered across compaction boundaries.

Inputs: session_id, raw_summary, facts_injected, injection_event, +3 more.

record_correctionCInspect

Record a user correction to Claude's output (high-priority learning signal)

Inputs: session_id (required), claude_action (required), user_correction (required), reasoning.

record_decisionCInspect

Record a decision trace: what the agent proposed and how the human responded

Inputs: session_id (required), decision_type (required), file_path, reasoning, +3 more.

record_eventCInspect

Record a development event (file edit, test run, etc.)

Inputs: event_type (required), session_id (required), description (required), success, +3 more.

record_test_outcomeCInspect

Record test results and link failures to recent code changes

Inputs: session_id (required), test_results (required).

resolve_contradictionCInspect

Pick a winner between two contradicting facts using a confidence-weighted strategy (auto, keep_higher_confidence, keep_most_recent, keep_most_sources, supersede_a, supersede_b, manual).

Inputs: fact_a_id (required), fact_b_id (required), notes, strategy.

search_globalCInspect

Search entities across all registered world-model projects

Inputs: query (required), limit.

seed_projectAInspect

Scan the project codebase and populate the knowledge graph with entities and relationships from existing code

Inputs: force, project_dir.

simulate_changeCInspect

Project blast radius and historical outcomes for a proposed change

Inputs: file_path (required), change_description (required).

validate_changeBInspect

Validate a proposed code change against known constraints

Inputs: file_path (required), change_type (required), proposed_content (required).

verify_retrievalAInspect

Adversarially verify an answer is grounded in a specific set of facts. An independent Coach LLM call checks each material claim in the answer against the supplied source facts and returns confidence (HIGH / MEDIUM / LOW), verified + unverified claim lists, and per-claim source_pointers. Never raises; failures return LOW + error populated. v0.12.12.

Inputs: query (required), answer (required), fact_ids (required), verification_model.

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    B3.2/5.0

    Scored across 31 tools

    Disambiguation4/5

    Most tools have clearly distinct purposes (e.g., query_fact vs. search_global, record_decision vs. record_correction). A few pairs like get_constraints/get_agents_md_constraints and simulate_change/validate_change have some overlap, but the descriptions differentiate them sufficiently for an agent to choose correctly.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern in snake_case (get_*, record_*, predict_*, etc.). Even compound names like get_co_edit_suggestions and prove_entry_inclusion maintain the pattern without mixing conventions.

    Tool Count2/5

    With 31 tools, the surface is significantly larger than the typical well-scoped server. While the server covers a broad domain, this count places a heavy burden on agent selection and navigation, likely reducing overall coherence.

    Completeness4/5

    The tool set covers a comprehensive lifecycle: seeding, ingestion, querying, validation, prediction, audit, and export. Minor gaps exist, such as no explicit update/delete for facts or constraints, but these are likely handled internally, so core workflows are supported without dead ends.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      Enables AI agents to sign decisions with post-quantum cryptographic proofs and maintain secure audit trails for compliance. It provides tools for stamping events, verifying chain integrity, and exporting audit data across industries like finance and healthcare.
      4
      55 npm
      MIT
    • F
      license
      Not graded
      quality
      F
      maintenance
      Superseded. Earlier version of the DCL product (MCP endpoint at /sse with API key, plus an offline desktop audit trail). Not the current path for agents. Use the hosted DCL Trust Oracle: https://github.com/Fronesis-Labs/dcl-webhook (MCP endpoint https://mcp.fronesislabs.com/mcp, Streamable HTTP, x402 payment, no API key).
      -
    • A
      license
      Not graded
      quality
      D
      maintenance
      Verifiable action receipts for AI agents — agents sign claims locally, an independent witness countersigns and timestamps, anyone can verify offline.
      9 npm
      MIT