dke-forge 0.25.0

CLI and MCP server for the DKE-Forge deterministic code intelligence engine

dke-forge

Python CLI and MCP server for DKE-Forge — a deterministic, graph-native code intelligence engine.

This package is a thin HTTP client over the Forge API. The reasoning engine itself (graph construction, pattern learning, provenance tracking, impact analysis) runs server-side at dke-forge.com. Your code is ingested into a private graph; every query returns grounded, provenance-tagged answers — no LLM guessing.

Install

pip install dke-forge              # CLI only
pip install 'dke-forge[mcp]'       # CLI + MCP stdio server

Configure

Sign up at dke-forge.com to get an API key, then:

dke-forge config set api_key fga_your_key

Or set environment variables (overrides the config file):

export FORGE_URL=https://dke-forge.com
export FORGE_API_KEY=fga_your_key

The config file lives at ~/.config/dke-forge/config.json (chmod 600).

CLI

dke-forge health                          # server status
dke-forge graph stats                     # node/edge counts
dke-forge graph function main             # look up a function
dke-forge graph callers authenticate      # who calls authenticate?
dke-forge graph impact login              # what breaks if login changes?
dke-forge scan /path/to/repo              # ingest a repo (polls to completion)
dke-forge scan /path/to/repo --no-wait    # returns scan_id; poll separately
dke-forge scan-status <scan_id>           # check async scan progress

Run dke-forge --help for the full command tree.

MCP — Claude Code, VS Code, Cursor, any MCP client

Install with the [mcp] extra, then add this to your MCP client config.

Claude Code (.mcp.json at project root, or ~/.mcp.json for global):

{
  "mcpServers": {
    "forge": {
      "command": "dke-forge-mcp",
      "env": {
        "FORGE_URL": "https://dke-forge.com",
        "FORGE_API_KEY": "fga_your_key"
      }
    }
  }
}

Cursor / VS Code (via settings.json):

{
  "mcp.servers": {
    "forge": {
      "command": "dke-forge-mcp",
      "env": {
        "FORGE_URL": "https://dke-forge.com",
        "FORGE_API_KEY": "fga_your_key"
      }
    }
  }
}

The server advertises 63 tools over stdio. Your assistant will discover them automatically and invoke them when it needs structural facts about your code.

63 Tools, 13 Categories

Category	Tools
Foundation	forge_graph_stats
Graphs	forge_list_graphs, forge_create_graph, forge_get_graph, forge_rename_graph, forge_delete_graph, forge_promote_default_graph
Navigation	forge_query_function, forge_callers_of, forge_callees_of, forge_depends_on, forge_impact_analysis, forge_callgraph_reachable, forge_find_implementers
Architecture	forge_dead_code, forge_subgraph, forge_circular_deps, forge_god_classes, forge_architecture_map, forge_api_surface, forge_abstractions, forge_module_topology
Change	forge_refactor_preview, forge_breaking_change, forge_deprecation_path, forge_risky_change, forge_commit_impact, forge_review
Rules	forge_rules, forge_check_rules, forge_declare_rule, forge_amend_rule, forge_retire_rule
Provenance	forge_contradictions, forge_last_modified, forge_provenance, forge_reasoning_trace
Patterns	forge_patterns, forge_pattern_catalog, forge_pattern_detail, forge_learn_patterns
Generation-prep	forge_envelope
Ingestion	forge_declare
Reasoning	forge_reason_chain, forge_audit_changes, forge_contradictions_hypothetical, forge_reach_from, forge_reach_cone, forge_use_def, forge_reaching_defs, forge_live_vars, forge_taint, forge_effects, forge_classify_effects, forge_effects_classification, forge_type_flow, forge_refine, forge_contract_check, forge_effect_behavioral
Meta	forge_health, forge_auth_verify
Invitations	forge_invitation_resend, forge_invitation_preview

Ingestion covers 8 languages: C++, C, Python, TypeScript, Go, Rust, Java, JavaScript. Cross-language relationships resolve into a single graph by construction — a Python call into a C++ library surfaces as a first-class edge.

Full tool reference: dke-forge.com/mcp.html

What Makes Forge Different

Every MCP tool answer is provenance-tagged — you can ask where a claim came from. Queries are deterministic — same graph, same answer. Contradictions in the graph are surfaced, not hidden. Pattern discovery has a lifecycle (proposed → supported → confirmed) with confidence propagation. This is the set of things a graph-native reasoning engine can do that LLM-plus-grep fundamentally cannot.

Requirements

• Python 3.8+
• A DKE-Forge account with an API key (sign up)

License

MIT. See LICENSE.

Links

• Homepage: dke-forge.com
• API Docs: dke-forge.com/docs.html
• MCP Guide: dke-forge.com/mcp.html