Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
KT_MCP_TOKENNoPre-shared access token used to authenticate the exact `/mcp/<access-token>` path for Streamable HTTP. Must contain 32–128 URL-safe characters (`A-Z`, `a-z`, `0-9`, `_`, or `-`). If unset, the desktop GUI generates a new token each launch and the raw HTTP server requires the token path for access.
TUNNEL_TOKENNoCloudflare Tunnel connector token used to prefill and start a fixed-domain tunnel in the desktop GUI. Passed to the connector via environment variable (not on the process command line).

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

Standalone KohakuTerrarium workspace tools. Relative paths use the active workspace configured for this server (shown in the desktop GUI when present). The operator can switch it without reconnecting; absolute paths in file results identify the workspace used for that call. Read files before modifying existing contents. Stale reads require rereading. File tools restrict paths to the workspace by default. Commands run with the user's permissions and are not sandboxed. Python uses a fresh process for each call. Background jobs have no automatic completion notification; poll job_status/job_wait. Jobs and read baselines are shared by clients of this instance and lost on restart. Do not replay a write or command automatically after losing its reply.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
readA

Read UTF-8 text with line numbers. offset is zero-based; limit defaults to 2000. Required before modifying an existing file.

writeA

Atomically create or replace UTF-8 text. Existing files require a prior read and unchanged contents.

editA

Exact text replacement after reading the file. An ambiguous match fails unless replace_all is true.

multi_editA

Apply ordered exact replacements to one file, committing once only if every replacement succeeds.

globA

Find files using shell-style matching on relative paths. Skips symlinks and common generated directories; does not parse .gitignore.

grepB

Search UTF-8 files with a Python regular expression. Returns matching lines; skips binary, oversized, and unreadable files.

treeC

List directory entries to a bounded depth. Skips symlinks and common generated directories.

bashA

Run Bash in the workspace. No sandbox. timeout defaults to 60 seconds (maximum 300). run_in_background returns a job_id; poll job_status/job_wait. Output is bounded.

pythonA

Run code in a fresh Python subprocess; variables do not persist. No sandbox. timeout defaults to 60 seconds (maximum 300). run_in_background returns a job_id; poll job_status/job_wait. Output is bounded.

job_statusA

Get a retained job or list jobs. Results are bounded in memory and lost on restart.

job_waitB

Wait up to timeout seconds (0–60, default 10). A wait timeout does not cancel the command.

job_cancelB

Cancel a running command and wait for process cleanup. Completed jobs are unchanged.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A3.6/5.0

Scored across 12 tools

Disambiguation4/5

Most tools target clearly distinct operations: file reading/writing, searching, shell execution, and job lifecycle management. The only meaningful overlap is edit versus multi_edit, but the descriptions make the single-replacement versus ordered-batch distinction clear enough.

Naming Consistency4/5

Core tools use short, conventional names like read, write, edit, glob, grep, bash, and python, while background-job tools consistently use the job_ prefix (job_status, job_wait, job_cancel). The mix of bare verbs/nouns and prefixed names is slightly uneven but still predictable and readable.

Tool Count5/5

Twelve tools is well-scoped for a coding workspace MCP: file editing, search, execution, and job management each have a small, justified set of operations. No tool appears redundant or out of place.

Completeness4/5

The surface covers the core coding-agent lifecycle: read/write/edit/multi_edit for files, glob/grep/tree for discovery, bash/python for execution, and job_status/job_wait/job_cancel for background processes. Minor gaps like a dedicated delete or move/copy tool exist, but agents can work around them via bash.

Maintenance

ActivityNo data
ResponsivenessOutdated data