Skip to content
Projects

Sandesh13fr/TCalc has 46 GitHub stars and 63 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-15 22:08 UTC. Last push 2026-11-13.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Update delayed
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

Local-first workspace token calculator and model recommender for coding agents, with repo maps, VS Code, CLI, and MCP.

Developer Tools

README.md

Original

TCalc measures local workspace tokens and turns them into model recommendations and agent-ready context

VS Code Marketplace Open VSX GitHub release TCalc website CI status MIT license

Scan a codebase once. See where its context goes, which model is enough for the job, and what your coding agent should read first.
Everything runs locally: no telemetry, cloud calls, or source upload.

Install · See it work · CLI · Coding agents · Website

See the workspace before your agent does

TCalc dashboard showing workspace token totals, model recommendations, and token-heavy files and folders

TCalc respects workspace ignore rules, estimates tokens by file, folder, and language, then recommends models against the actual context and your current goal.

Compare models against the scan Generate a budgeted repo map
TCalc model comparison with fit, context, and estimated cost A TCalc repo map generated within a token budget

One scan, three useful decisions

Decision TCalc produces
01 What is consuming context? Included and excluded files, token-heavy paths, folders, languages, and warnings.
02 Which model is enough? Cheapest-sufficient, balanced, and high-confidence options with context fit and estimated cost.
03 What should the agent read? Budgeted repo maps, goal-aware agent rules, reports, and ready-to-use MCP configuration.

The scanner, recommender, report generator, and integrations share the same local-first core. Model catalogs stay provider-neutral, and recommendations change with the task rather than hard-coding one model as “best.”

Start in VS Code

Install TCalc from the VS Code Marketplace or Open VSX, then run:

Command Palette → TCalc: Quick Start

The guided flow scans the open workspace, sets a goal, compares models, and opens the dashboard. You can also install a downloaded release directly:

code --install-extension tcalc-*.vsix

See the VSIX installation guide for the local package workflow. A source-built JetBrains plugin is also available as an MVP for IntelliJ-based IDEs.

Use the CLI

Build once, then scan, compare, or generate only the context you need:

Building from source needs Node.js 22 and pnpm 11.5.2. For the fork, clone, and setup walkthrough, see CONTRIBUTING.md.

pnpm install
pnpm build

pnpm cli scan ./my-project
pnpm cli recommend ./my-project --goal build-mvp
pnpm cli repo-map ./my-project --budget 8000
pnpm cli rules ./my-project --target codex --mode repo-map-first

Other commands generate Markdown/JSON reports, validate model catalogs, create MCP configs, and start the local MCP server. Run pnpm cli --help for the complete command list.

Ignore sources and precedence

TCalc applies ignore patterns in this order, with later matching rules taking precedence under normal gitignore semantics:

  1. root .gitignore;
  2. root .cursorignore;
  3. root .aiderignore;
  4. root .continueignore;
  5. root .tcalcignore;
  6. any explicit additional ignore files supplied by an integration;
  7. nested .gitignore files, scoped to their containing directory;
  8. workspace exclude patterns from TCalc configuration.

Nested ignore discovery skips version-control metadata and directories already excluded by higher-level rules. As with Git, a file cannot be re-included when its parent directory remains ignored. File-size, binary, generated-file, secret, and other safety exclusions are applied separately after path ignores and cannot be weakened by a negated ignore rule.

Choose a privacy mode

TCalc applies the same privacy setting in its CLI, VS Code UI, and shared configuration:

Mode Recommendation guarantee
local-first (default) Only catalog entries with supportsLocal: true are eligible. Cloud models and hybrid entries without a verified local execution path are rejected. If none qualify, TCalc reports that no eligible model is available instead of silently falling back to cloud execution.
cloud-ok Cloud, hybrid, and local catalog entries may be recommended according to fit, quality, and cost. This expands recommendation eligibility; it does not upload source code or invoke a provider by itself.

Set privacyMode in the workspace's .tcalc.json or use the corresponding privacy selector exposed by the CLI/VS Code integration. For example:

{
  "privacyMode": "local-first"
}

Workspace scanning, token estimation, repo-map generation, and reports remain local in both modes. Selecting cloud-ok only permits cloud-capable models to appear in recommendations; any later provider execution and credentials are controlled by the user's chosen external tool.

Connect your coding agent

TCalc can expose the same scan, recommendation, repo-map, rules, and report capabilities over stdio MCP. It also generates agent rule files and client configuration without uploading source code.

Tool Integration Guide
Cursor MCP + .cursorrules Setup
Continue MCP Setup
Claude Desktop MCP Setup
Claude Code MCP + CLAUDE.md Setup
Cline MCP Setup
Roo MCP Setup
Any stdio MCP client MCP Generic setup
# Generate a client config
pnpm cli mcp-config --target cursor
pnpm cli mcp-config --target continue --output continue-mcp.yaml

# Start the experimental local server
pnpm cli mcp
Goal-aware AGENTS.md rules generated by TCalc for a scanned workspace

[!NOTE] The MCP server is an experimental MVP: stdio transport only, with no hosted mode. Repo maps are structural and do not expose full source file bodies. See the MCP documentation for tools, resources, prompts, security boundaries, and current limits.

Repository map

Packages and applications
Package Role
@wma/core Shared types, configuration, and constants
@wma/scanner Workspace discovery, ignores, and classification
@wma/tokenizers Heuristic and provider-specific token estimation
@wma/model-catalog Model metadata and catalog validation
@wma/recommender Goal-aware fit, scoring, and cost estimates
@wma/repo-map Context-budgeted structural maps and symbols
@wma/agent-rules Rules for Codex, Claude Code, Cursor, and other agents
@wma/reports Markdown and JSON reports
@wma/mcp-server Local stdio MCP tools, resources, and prompts
apps/cli Commander-based CLI
apps/vscode-extension VS Code dashboard and commands
apps/jetbrains-plugin IntelliJ Platform plugin MVP
apps/dashboard Next.js website, documentation, local report viewer, and optional report service

Project guides

First contribution? Start here

First-timers-only friendly

Browse available beginner issues and read the first-contribution walkthrough. Tasks labelled first-timers-only are reserved for your first open-source PR. Comment to request assignment before starting; maintainers can help with setup and review feedback.

Troubleshooting VS Code test output

License

MIT