Atlas / MCP servers / heratiki / LocaLLama

LocaLLamaBLOCK

Name: LocaLLama
Author: heratiki

mcp/heratiki/locallama

MCP server by heratiki · clinebot · mcp server · roocode · Audited 2026-10-08

An MCP Server that works with Roo Code/Cline.Bot/Claude Desktop to optimize costs by intelligently routing coding tasks between local LLMs free APIs and paid APIs.

Verdict

BLOCK

Grade

Trust score

53 /100

Exposed tools

20 16r · 3w · 1d

Transport

stdio

License

—

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

[](https://github.com/Heratiki/locallama-mcp/releases) [](https://github.com/Heratiki/locallama-mcp/releases)

Local-first, provider-neutral Model Context Protocol server for coding-agent workflows. Routes tasks across local models (Ollama, LM Studio, llama.cpp), free OpenRouter models, and paid frontier models using cost, latency, context capacity, and benchmark history.

Node.js: >=22

⚠️ Early / experimental — not yet a stable release. This project is under active, rapid development and has not been fully verified end-to-end. MCP tool signatures, configuration, and behavior may change between releases without notice. Version numbers follow SemVer mechanically (they're derived from Conventional Commit messages, not hand-picked), so a 1.x number signals only "a public surface exists" — it is not a promise of stability or completeness. If you depend on this server, pin to an exact version. - Tagged releases on `main` are the relatively safer builds. - The `testing` channel publishes bleeding-edge pre-releases (x.y.z-testing.n) for trying unproven changes early.

Overview

LocalLama MCP reduces token costs without sacrificing quality. Tasks are queued asynchronously — route_task returns a task_id immediately; callers poll get_task_status for results. The decision engine chooses local → free → paid based on measured provider capabilities and configurable thresholds.

Supported MCP clients: Codex, Claude Code, Claw Code, Cursor, GitHub Copilot Agent mode, and any generic MCP stdio client.

Requirements

Node.js 22+
npm
At least one of: Ollama, LM Studio, llama.cpp server, or an OpenRouter API key

Read from source at commit 9c03639448efOBSERVED · 2026-10-08

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add locallama-mcp --env OPENROUTER_API_KEY=${OPENROUTER_API_KEY} --env TOKEN_THRESHOLD=${TOKEN_THRESHOLD} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "locallama-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "OPENROUTER_API_KEY": "${OPENROUTER_API_KEY}",
        "TOKEN_THRESHOLD": "${TOKEN_THRESHOLD}"
      }
    }
  }
}

Exposed tools (20)

16 read · 3 write · 1 destructive. Blast radius: 1 tool can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`benchmark_free_models`	read	Benchmark the performance of free models from OpenRouter
`benchmark_model`	write	Run built-in benchmark suites against a specific model identified by its model id.
`benchmark_task`	read	Benchmark the performance of local LLMs vs paid APIs for a specific task. When live job monitoring is available, the response includes monitoring.websocketUrl and MCP job resource URIs.
`benchmark_tasks`	read	Benchmark the performance of local LLMs vs paid APIs for multiple tasks. When live job monitoring is available, the response includes monitoring.websocketUrl and MCP job resource URIs.
`cancel_job`	read	Cancel a running background job. Returns the job id, whether cancellation succeeded, and the final status.
`cancel_task`	read	Cancel all queued or in-progress jobs belonging to a non-blocking route_task submission.
`check_for_updates`	read	Check whether the running locallama-mcp server is up to date with the latest commit on the future-testing branch on GitHub. Returns upToDate status, local SHA, remote SHA, and any error.
`clear_openrouter_tracking`	destructive	Clear OpenRouter tracking data and force an update
`get_cost_estimate`	read	Returns a structured JSON cost estimate for running a given number of tokens through the router.
`get_free_models`	read	Get a list of free models available from OpenRouter
`get_system_state`	read	Return a structured snapshot of current runtime health: local slot occupancy (inference vs benchmark vs idle),
`get_task_status`	read	Poll a non-blocking route_task submission. Returns aggregate task status, progress, retry hint, and per-job summaries including inline result content when completed.
`get_telemetry_summary`	read	Return an analytics summary of LLM call telemetry recorded across all runs.
`preemptive_route_task`	read	Cheap model-selection check that returns a routing recommendation WITHOUT executing the task.
`rate_model`	read	Provide explicit human or orchestrator feedback on generator or validator model performance for a completed task/job. Updates capability scores and enqueues fixture candidates for negative outcomes.
`reload_config`	write	Reload runtime configuration from the project .env file and apply only hot-reloadable fields.
`retriv_init`	read	Initialize and configure Retriv for code search and indexing
`retriv_search`	read	Search code using native BM25 engine
`route_task`	read	Queue a coding task for execution by the most cost-effective LLM available and return immediately with a task_id.
`set_model_prompting_strategy`	write	Update the prompting strategy used for a specific OpenRouter model. Allows callers to supply a custom system prompt, user prompt template, and assistant prompt template that will be used on subsequent task executions for that model.

Trust audit

BLOCKgrade F · trust 53/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	FAIL
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (6 observation(s))
Network: declared (2 observation(s))
Shell: declared (5 observation(s))
Dependencies: not all pinned
Secrets in source: none-found

Findings (20)

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/modules/decision-engine/services/codeValidator.ts:46

const containsDangerousPatterns = code.includes('eval(') ||

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/modules/decision-engine/services/codeValidator.ts:47

code.includes('new Function(') ||

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

src/modules/llama-cpp/discovery.ts:21

exec(cmd, (error, stdout) => {

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHPrompt injection · prompt.override · CWE-94, CWE-1427

docs/PLAN.md:1108

- Attempt to override system-prompt instructions in the local model (jailbreak attempts).

Why it matters. asks the agent to drop prior instructions or safety

Fix. remove the instruction

MEDIUMFilesystem / path · mcp.destructive_tools · CWE-22, CWE-59

clear_openrouter_tracking

Why it matters. 1 tool(s) can delete or overwrite

Fix. prefer a read-only mode or scoped tokens; the page states the blast radius

LOWInventory / provenance · inv.hidden_file · CWE-1104

.releaserc.json

.releaserc.json

Why it matters. hidden member outside the usual dotfiles

Fix. review its purpose

LOWInventory / provenance · inv.hidden_file · CWE-1104

.rooignore

.rooignore

Why it matters. hidden member outside the usual dotfiles

Fix. review its purpose

LOWInventory / provenance · inv.no_license · CWE-1104

Why it matters. no LICENSE file and no repo licence

Fix. add a licence

LOWInsecure crypto · crypto.weak_hash · CWE-327, CWE-338

src/modules/benchmark/storage/results.ts:47

const testHash = crypto.createHash('md5')

LOWFilesystem / path · fs.traversal · CWE-22, CWE-59

src/modules/api-integration/cost-estimation/index.ts:1

import { costMonitor } from '../../cost-monitor/index.js';

LOWFilesystem / path · fs.traversal · CWE-22, CWE-59

src/modules/api-integration/cost-estimation/index.ts:2

import { openRouterModule } from '../../openrouter/index.js';

LOWFilesystem / path · fs.traversal · CWE-22, CWE-59

src/modules/api-integration/cost-estimation/index.ts:4

import { config } from '../../../config/index.js';

LOWFilesystem / path · fs.traversal · CWE-22, CWE-59

src/modules/api-integration/cost-estimation/index.ts:5

import { logger } from '../../../utils/logger.js';

LOWFilesystem / path · fs.traversal · CWE-22, CWE-59

src/modules/api-integration/openrouter-integration/index.ts:1

import { config } from '../../../config/index.js';