Skip to main content

promptbridge

Type coding requests in your own language. Get a clear English task spec your AI coding agent can execute without guessing.

ภาษาไทย · Changelog · Contributing

Many developers read English fine but don't think in it fast enough to write precise prompts. promptbridge sits between you and your coding agent (Claude Code, Claude Desktop, Cursor, any MCP client):

/spec หน้า login อยากให้ถ้าใส่รหัสผิดหลายครั้งมันล็อคไว้ก่อน
      (login page: lock it after too many wrong passwords)
  1. It asks only what is ambiguous — at most 3 tap-to-answer questions, in your language: how many attempts? lock for how long? per account or per IP?
  2. It shows a 2–3 line summary in your language plus the English spec: Goal · Context · Requirements · Constraints · Acceptance criteria · Assumptions.
  3. You approve, and the agent does the work — reporting back in your language, with code, paths and errors left untouched.

It also learns your vocabulary: tell it once that «ตะกร้า» means CartStore, and every future spec uses the real name.

How it works

Part Does
Host model (the Claude/LLM you already use) Reads your language, fills spec slots in English, asks the questions, summarizes results
promptbridge MCP server (Python) Task templates, decides which questions matter, renders the spec deterministically, remembers your glossary and specs that worked

The server never calls an LLM itself — no extra API cost, no data leaves your machine. Everything is stored locally in ~/.promptbridge/.

Install

You need uv.

Claude Code

# 1) Register the MCP server for all projects
claude mcp add promptbridge -s user -- uvx promptbridge-mcp

# 2) Install the /spec skill
mkdir -p ~/.claude/skills/spec
curl -fsSL https://raw.githubusercontent.com/Endistic/promptbridge/main/skills/spec/SKILL.md \
  -o ~/.claude/skills/spec/SKILL.md

# 3) In any project
/spec ปุ่มบันทึกในหน้าโปรไฟล์กดแล้วขึ้น error 500

Claude Desktop

Claude menu (macOS menu bar) → Settings… → Developer → Edit Config, add the server, then quit (⌘Q) and reopen:

{
  "mcpServers": {
    "promptbridge": { "command": "/opt/homebrew/bin/uvx", "args": ["promptbridge-mcp"] }
  }
}

Use the full path from which uvx — Claude Desktop does not see your shell's PATH. Then ask: "Use promptbridge to make a spec for: …"

Cursor and other MCP clients

{ "mcpServers": { "promptbridge": { "command": "uvx", "args": ["promptbridge-mcp"] } } }

Clients without the skill follow the workflow described in the tools themselves.

From source

git clone https://github.com/Endistic/promptbridge
claude mcp add promptbridge -s user -- uvx --from ./promptbridge promptbridge

Or load it as a Claude Code plugin (server + skill): claude --plugin-dir ./promptbridge, then /promptbridge:spec.

Usage

Type Result
/spec <request> Full flow: ask → summarize → do
/spec --quick <request> Skip questions; unknowns become visible Assumptions in the spec
"remember: ตะกร้า = CartStore" Add a glossary term
"show my glossary" List glossary terms
"have I asked for something like this before?" Search specs that worked

Task types

bug_fix · feature · refactor · test · explain — each has its own required facts (src/promptbridge/templates/). Adding a type is one YAML file.

Glossary

  • personal — ~/.promptbridge/pb.db, follows you across projects
  • repo — <repo>/.promptbridge/glossary.yaml, commit it so the whole team shares one vocabulary

When you edit a spec and add a real name (e.g. LoginAttemptService), promptbridge offers to remember the pairing.

Languages

Built around a locale from day one. Thai and English have hand-written question phrasing; any other language (Japanese, Vietnamese, Indonesian, Korean, Spanish, …) works today — the host model translates the questions. Adding native phrasing for a language is one file: src/promptbridge/locales/. See CONTRIBUTING.md.

MCP tools

Tool Purpose
pb_capture Start from the user's raw request: guess task type, match glossary, scan the repo
pb_clarify Decide which questions to ask (max 3 per round, max 2 rounds)
pb_compile Render the English spec; unconfirmed guesses become Assumptions
pb_save Record accepted / edited / rejected; suggest glossary terms from edits
pb_glossary List / add / remove terms (personal or repo scope)
pb_library Search specs that worked before

Resource promptbridge://stats reports accepted / edited / rejected counts.

Configuration

Variable Default Meaning
PROMPTBRIDGE_HOME ~/.promptbridge Where the local database lives

Works with MCP Python SDK 1.x and 2.x.

License

MIT

Metadata

Release files for promptbridge-mcp 0.2.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for promptbridge-mcp 0.2.0
File Size Uploaded
promptbridge_mcp-0.2.0.tar.gz 32.4 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for promptbridge-mcp 0.2.0
File Interpreter ABI Platform
promptbridge_mcp-0.2.0-py3-none-any.whl Python 3 none any Details

Total release size: 62.7 kB

Release files / promptbridge_mcp-0.2.0.tar.gz

Download URL promptbridge_mcp-0.2.0.tar.gz
Size 32.4 kB
Tags Source
SHA-256 checksum
How to use checksums
244e9f1464d60689718cfa2f585974c4c6d54f0a30ea13bd9caa84288ea955f7
BLAKE2b-256 checksum
How to use checksums
beab2c8091d1b091b4690762d6c356d230ae6aae9c9c29c3d282fb2f4cae8c57
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 30, 2026.

Transparency log

Release files / promptbridge_mcp-0.2.0-py3-none-any.whl

Download URL promptbridge_mcp-0.2.0-py3-none-any.whl
Size 30.3 kB
Tags Python 3
SHA-256 checksum
How to use checksums
66ac4554a45f2333c73444010fbc7c1214e641113164207e3271b50aabc6c1af
BLAKE2b-256 checksum
How to use checksums
b6f5638b1b4de656e922df950db7490cda5d80d0ed4032a5e13e3a9f7ea86ffc
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 30, 2026.

Transparency log

Release history Release notifications | RSS feed

This release

0.2.0 This release

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page