@dtst/txt

MCP server and CLI for writing text through any OpenAI-compatible API. It reads files, globs, directories, URLs or earlier results as context, and can look at one or many images.

npx -y @dtst/txt    # MCP stdio server

Tools: write_text, chat, read_context, list_models.

Configuration

OPENAI_BASE_URL=https://openrouter.ai/api/v1   # any OpenAI-compatible root
OPENAI_API_KEY=sk-...
OPENAI_MODEL=openai/gpt-6-luna                 # a TEXT model

Optional: DEBUG=true for verbose stderr logs, DTST_WORKSPACE=/path to resolve relative paths, DTST_OUTPUT_DIR=out for a default destination. A .env file next to your project is picked up automatically.

Add to your client

This entry works as-is in Claude Desktop, Cursor, Windsurf, Cline and Roo Code:

{
  "mcpServers": {
    "txt": {
      "command": "npx",
      "args": ["-y", "@dtst/txt"],
      "env": {
        "OPENAI_BASE_URL": "https://openrouter.ai/api/v1",
        "OPENAI_API_KEY": "sk-...",
        "OPENAI_MODEL": "openai/gpt-6-luna"
      }
    }
  }
}

| Client | Where it goes | Key | | ----------------- | ---------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------ | | Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json (macOS), %APPDATA%\Claude\claude_desktop_config.json (Windows) | mcpServers | | Cursor | ~/.cursor/mcp.json or .cursor/mcp.json | mcpServers (add "type": "stdio") | | Windsurf | ~/.codeium/windsurf/mcp_config.json | mcpServers | | Cline | ~/.cline/mcp.json | mcpServers | | Roo Code | .roo/mcp.json | mcpServers | | VS Code (Copilot) | .vscode/mcp.json | servers + "type": "stdio" | | Zed | Zed settings.json | context_servers | | JetBrains | Settings → Tools → AI Assistant → MCP → Add → STDIO | paste the JSON above |

Use a text model: openai/gpt-6-luna, anthropic/claude-opus-5.5, qwen/qwen3.8-max. txt models --vision lists models that accept images.

opencode

opencode.json — command is an array and env vars use environment:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "txt": {
      "type": "local",
      "command": ["npx", "-y", "@dtst/txt"],
      "environment": {
        "OPENAI_BASE_URL": "https://openrouter.ai/api/v1",
        "OPENAI_API_KEY": "sk-...",
        "OPENAI_MODEL": "openai/gpt-6-luna"
      },
      "enabled": true
    }
  }
}

Codex CLI

~/.codex/config.toml:

[mcp_servers.txt]
command = "npx"
args = ["-y", "@dtst/txt"]

[mcp_servers.txt.env]
OPENAI_BASE_URL = "https://openrouter.ai/api/v1"
OPENAI_API_KEY = "sk-..."
OPENAI_MODEL = "openai/gpt-6-luna"

Command-line clients

claude mcp add --transport stdio txt \
  --env OPENAI_API_KEY=sk-... --env OPENAI_MODEL=openai/gpt-6-luna \
  -- npx -y @dtst/txt

codex mcp add txt --env OPENAI_MODEL=openai/gpt-6-luna -- npx -y @dtst/txt

gemini mcp add -e OPENAI_MODEL=openai/gpt-6-luna txt npx -y @dtst/txt

Tools

write_text

One-shot writing: draft, summarise, rewrite, translate, extract.

| param | type | what it does | | ------------------------------------------------------------ | ------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ | | instructions | required, string | The task, e.g. Summarise these release notes as 5 bullets for a non-technical reader. Placed last in the prompt, which models follow best. | | input | string | Inline source material to transform (short texts). | | context | object | Reference material: { files, dirs, urls, text, artifacts }. | | context.files | string[] | Paths, globs, file:// URIs or dtst://artifact/<id>. | | context.dirs | string[] | Directories (non-recursive, capped at 64 files). | | context.urls | string[] | http(s) URLs returning text, JSON or an image. | | context.text | string[] | Inline snippets treated as context. | | context.artifacts | string[] | Artifacts from earlier calls. | | images | string[] ≤16 | Images for vision models: paths, globs, URLs, data URLs, base64, artifact URIs. | | model | string | One-off override of OPENAI_MODEL. | | system | string | Extra system guidance (persona, audience, constraints). | | max_tokens | int 1–200000 | Output budget. Raise it for reasoning models. | | temperature | number 0–2 | Sampling temperature. | | top_p | number 0–1 | Nucleus sampling. | | stop | string[] ≤4 | Stop sequences. | | reasoning_effort | string | low, medium, high for models that support it. | | format | text | markdown | json | json switches on JSON mode; the content is not validated for you. | | verbosity | concise | balanced | detailed | How much detail to produce. | | stream | boolean | Stream from the provider and report progress. Default false (most compatible). | | output_path, output_dir, filename, overwrite, save | | Saving controls (see below). Text is returned inline either way. |

chat

Multi-turn conversation; stateless, so send the whole history.

| param | type | what it does | | -------------------------------------------------------------------------------------------------------------------- | --------------- | ---------------------------------------------------------------------------------------------------------------- | | messages | required, array | Oldest first: [{ "role": "user" \| "assistant" \| "system", "content": "…" }]. The last user turn is answered. | | images | string[] ≤16 | Attached to the last user turn. | | model, system, max_tokens, temperature, top_p, stop, reasoning_effort, format, verbosity, stream | | Same as write_text. | | output_path, output_dir, filename, overwrite, save | | Same saving controls. |

read_context

Brings reference material into the conversation as text + image content.

| param | type | what it does | | ---------------------- | ---------------- | ----------------------------------------------------------------------------- | | files | string[] | Paths, globs, file:// URIs, artifact URIs. | | dirs | string[] | Directories (non-recursive, ≤64 files). | | urls | string[] | http(s) URLs returning text, JSON or an image. | | text | string[] | Inline snippets passed through. | | images | string[] ≤16 | Image sources to load as image content. | | max_bytes_per_source | int 1024–8388608 | Per-source budget (default 262144); longer files are truncated with a marker. | | include_content | boolean | false returns only sizes and paths. Default true. |

list_models

| param | type | what it does | | ------- | ---------- | ---------------------------------------- | | query | string | Substring filter over model id and name. | | limit | int 1–1000 | Maximum results (default 100). |

How an agent uses it

read_context { "files": ["docs/*.md"], "max_bytes_per_source": 65536 }

write_text {
  "instructions": "Summarise what changed in 3.2 as 5 bullets, then list breaking changes",
  "context": { "files": ["docs/release-3.2.md"] },
  "verbosity": "concise", "format": "markdown",
  "output_path": "reports/release-summary.md"
}

chat {
  "messages": [
    { "role": "user", "content": "What does this screenshot show?" },
    { "role": "assistant", "content": "A stack trace from the worker pool." },
    { "role": "user", "content": "Which line is the root cause?" }
  ],
  "images": ["shots/error.png"]
}

Context reads are budgeted per source and in total, and every block is fenced and labelled as untrusted data so file contents cannot smuggle in instructions. Anything skipped is reported back instead of being dropped.

Saving to a local file

Text is returned inline; it is also written when you give it a destination. output_path wins over output_dir; filename sets the name and the extension comes from format (.md, .json). Relative paths resolve against DTST_WORKSPACE (default: the directory the server was started in), or DTST_OUTPUT_DIR when set. overwrite: false (default) appends -1, -2, … instead of clobbering; save: true forces a write even without a destination.

write_text { "instructions": "…", "output_path": "reports/summary.md" }
write_text { "instructions": "…", "format": "json", "output_dir": "data", "filename": "extract" }
chat       { "messages": [ … ], "save": true, "filename": "reply.md" }

Writes are atomic, and each result carries the absolute path, a file:// resource link and a dtst://artifact/<id> URI that later calls in the same session can use as context.artifacts.

CLI

txt write "Summarise RELEASE.md in five bullets" --context RELEASE.md
txt write "Describe these screenshots" -i "shots/*.png" --format markdown
txt chat "Explain this stack trace" -i screenshot.png
txt read src/*.ts --json
txt models --vision
txt --help

MIT licensed.