Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Llama-Bridge Implement a recursive Fibonacci function in Python"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server that lets your cloud model (Gemini / Claude) delegate implementation work to a local llama.cpp server, preserving precious cloud-model usage limits while maintaining high code quality through AI-powered code review.
Cloud Model → Plans & Reviews
↕ MCP
Local Model → Writes CodePython 3.11+
uv (recommended) or pip
A running llama.cpp server (see below)
# Example with llama-server
./llama-server -m your-model.gguf --port 8080
# Or with llama-cpp-python
pip install llama-cpp-python[server]
python -m llama_cpp.server --model your-model.gguf --port 8080We provide an automated installer script that creates a virtual environment, installs the package and its dependencies, and automatically configures llama-bridge in your global Antigravity/Gemini configuration directory (~/.gemini/config/mcp_config.json on Linux/macOS):
python install.pyTo enable the cloud model to automatically use the local Llama-Bridge delegation tools across all workspaces:
Open the project-scoped .agents/AGENTS.md file.
Copy its entire content.
Paste the content into your global GEMINI.md instructions file located at ~/.gemini/GEMINI.md.
The installer will set the default local server URL to http://localhost:8080. If you need to customize this, or set a custom API key, you can add environment variables to the "env" block in your global mcp_config.json or create a .env file in the project directory:
# Example .env settings:
LLAMA_BASE_URL=http://localhost:8080
LLAMA_REQUEST_TIMEOUT=120Restart Antigravity IDE. The cloud model should now have access to:
implement_code
generate_tests
refactor_code
fix_code
generate_docs
check_local_model_health
Related MCP server: local-llm-mcp
Tool | Purpose | Inputs |
implement_code | Generate implementation from a spec | task_description, language, context, constraints |
generate_tests | Generate test code | code, language, framework, requirements |
refactor_code | Apply a specific refactor | code, language, refactor_description, constraints |
fix_code | Fix bugs from errors/feedback | code, language, errors, review_comments |
generate_docs | Generate documentation | code, language, style |
check_local_model_health | Check server availability | (none) |
Every code tool returns a consistent ToolResponse:
{
"success": true,
"code": "def hello(): ...",
"error": null,
"metadata": {
"tool": "implement_code",
"elapsed_seconds": 3.42,
"usage": {"prompt_tokens": 150, "completion_tokens": 89},
"warnings": []
}
}All settings are configured via environment variables or a .env file:
Variable | Default | Description |
|
| llama.cpp server URL |
|
| Timeout in seconds |
|
| Default sampling temperature |
|
| Default token budget |
|
| Model identifier (usually ignored) |
uv run pytest tests/ -vThe cloud model (Gemini/Claude in Antigravity IDE) acts as a senior engineer — it plans, delegates, and reviews. The local model acts as a fast junior engineer — it writes code quickly. The MCP server is the bridge between them.
Cloud model receives a user request
Cloud model breaks it into implementation tasks
Cloud model calls MCP tools to delegate coding
Local model generates implementation
Cloud model reviews the code
If issues found → calls fix_code with feedback
Repeat until code meets quality standards
Cloud model presents the final, reviewed code
This gives you practically unlimited coding capacity from the local model, with cloud-grade quality assurance from the review loop.
Check if the local llama.cpp server is reachable and loaded.
Returns availability status and model metadata. Call this before starting a batch of implementation tasks, or when a previous tool call failed with a connection error, to decide whether to retry or fall back to direct cloud-model implementation.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
| Name | Required | Description |
|---|---|---|
| error | No | Error message if the server is unreachable. |
| model | No | Model identifier reported by the server, if available. |
| available | Yes | Whether the local llama.cpp server is reachable and loaded. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description aligns with readOnlyHint=true by describing a check rather than a mutation, and it adds useful behavioral context by saying it returns availability status and model metadata. This is transparent for a read-only health-check tool with no side effects to disclose.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three short sentences with no filler. The core purpose comes first, the return summary second, and the usage guidance third. Every sentence adds distinct value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a zero-parameter read-only health check that already has an output schema, the description covers purpose, returned information, and when to invoke it. There is no missing detail an agent would need to call it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
There are zero parameters, and the schema coverage is 100%, so the description has no parameter documentation burden. The no-parameter baseline of 4 applies because there is nothing missing or ambiguous to explain.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description names the exact verb and resource: 'Check if the local llama.cpp server is reachable and loaded.' It also states the return value ('availability status and model metadata'), which makes the tool's purpose unmistakable and distinguishes it from the sibling implementation tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides clear trigger conditions: call before a batch of implementation tasks, or after a previous tool call fails with a connection error. It also explains the decision being made ('whether to retry or fall back to direct cloud-model implementation'). It does not explicitly state when not to use it, so it falls just short of a 5.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Fix bugs in code based on errors or review feedback.
Provide the broken code along with compiler/runtime errors and/or reviewer comments. The local model will return a corrected version that addresses all reported issues.
IMPORTANT: Always review the returned fix before accepting it.
| Name | Required | Description | Default |
|---|---|---|---|
| code | Yes | The code that contains bugs or issues. | |
| errors | No | Compiler errors, runtime exceptions, or test failure output. Paste the exact error messages. | |
| language | Yes | Programming language of the code. | |
| max_tokens | No | Max tokens override. | |
| temperature | No | Temperature override. | |
| review_comments | No | Code review feedback describing what's wrong and what needs to change. Be specific about the issues. |
| Name | Required | Description |
|---|---|---|
| code | No | The generated artefact (code, tests, documentation). None on failure. |
| error | No | Human-readable error message. None on success. |
| success | Yes | Whether the generation completed successfully. |
| metadata | No | Auxiliary data: token usage, elapsed time, quality warnings, etc. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations already provide readOnlyHint=true, and the description adds useful behavioral context by noting that the local model 'will return a corrected version' and that the agent should 'Always review the returned fix before accepting it.' This warns about the model's fallibility without contradicting the read-only annotation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three short, purposeful paragraphs: what the tool does, what inputs to provide, and an important caution. The 'IMPORTANT' warning earns its place and is not excessive. There is no redundant filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's simplicity, its output schema availability, and the annotations, the description is complete. It specifies required inputs, the expected output behavior, and a review caution. Nothing essential is missing for an agent deciding whether and how to call this tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, and the schema already describes code, errors, language, review_comments, and optional overrides. The description echoes the input requirements but does not add new parameter-level details beyond what the schema provides, so the baseline score of 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description opens with a specific verb and resource: 'Fix bugs in code based on errors or review feedback.' This clearly distinguishes fix_code from siblings like refactor_code (which implies improving structure rather than fixing bugs) and implement_code (which implies creating new code). The purpose is unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It provides clear usage context: 'Provide the broken code along with compiler/runtime errors and/or reviewer comments.' It tells the agent what inputs are needed and what will be returned. However, it does not explicitly name sibling tools or state when not to use it, so it stops short of full alternative routing.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Generate documentation for existing code.
Produces documentation in the requested style — inline docstrings, README content, or structured API reference.
IMPORTANT: Always review the returned documentation for accuracy.
| Name | Required | Description | Default |
|---|---|---|---|
| code | Yes | The code to generate documentation for. | |
| style | No | Documentation style: 'docstring' for inline docs, 'readme' for a README section, 'api_reference' for structured API docs. | docstring |
| language | Yes | Programming language of the code. | |
| max_tokens | No | Max tokens override. | |
| temperature | No | Temperature override. |
| Name | Required | Description |
|---|---|---|
| code | No | The generated artefact (code, tests, documentation). None on failure. |
| error | No | Human-readable error message. None on success. |
| success | Yes | Whether the generation completed successfully. |
| metadata | No | Auxiliary data: token usage, elapsed time, quality warnings, etc. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already mark the tool as readOnly, and the description adds a genuinely useful behavioral warning: 'Always review the returned documentation for accuracy.' This warns the agent that generated docs may be imperfect, which is valuable beyond the structured annotation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Three short, front-loaded sentences with no filler. The purpose comes first, the style options follow, and the important accuracy warning is isolated for emphasis.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the output schema, full schema coverage, and readOnly annotation, the description is complete. It establishes the input (existing code), the style choices, and the need for output review, leaving no critical gap for an agent to call it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the schema carries the full parameter documentation. The description summarizes the style enum values, but this largely duplicates the schema rather than adding new semantic detail.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource: 'Generate documentation for existing code.' It names the distinct deliverables (docstrings, README, API reference), which clearly separates it from sibling tools like generate_tests, implement_code, or refactor_code.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives clear context—use it for existing code and choose a documentation style—but it never explicitly states when to prefer this tool over siblings like generate_tests or refactor_code. There are no exclusions or alternative tool mentions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Generate test code for a given implementation.
Produces a complete, runnable test file that covers happy paths and edge cases. Specify the testing framework and any particular scenarios you want covered.
IMPORTANT: Always review the returned tests before accepting them.
| Name | Required | Description | Default |
|---|---|---|---|
| code | Yes | The implementation code to generate tests for. | |
| language | Yes | Programming language of the implementation. | |
| framework | No | Testing framework to use (e.g. 'pytest', 'jest', 'go test', 'junit'). Leave empty for the language's default. | |
| max_tokens | No | Max tokens override. | |
| temperature | No | Temperature override. | |
| requirements | No | Specific scenarios, edge cases, or coverage requirements for the generated tests. |
| Name | Required | Description |
|---|---|---|
| code | No | The generated artefact (code, tests, documentation). None on failure. |
| error | No | Human-readable error message. None on success. |
| success | Yes | Whether the generation completed successfully. |
| metadata | No | Auxiliary data: token usage, elapsed time, quality warnings, etc. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations provide readOnlyHint=true, and the description does not contradict this—generating test code produces output without mutating state. The description adds value beyond annotations by disclosing that the output is a complete, runnable file and by warning to always review returned tests before accepting, implying outputs may be imperfect. This is useful behavioral context.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact—two concise sentences plus an IMPORTANT warning—with every sentence earning its place: it states the action, describes the output, guides input configurability, and advises on output review. No redundancy or irrelevant detail.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a generative tool with 6 parameters and an output schema, the description adequately covers purpose, output nature, and optional input guidance. The review warning addresses output reliability risk. It doesn't discuss all potential failure modes, but full schema coverage and the output schema mitigate that gap, making the definition sufficient for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the baseline is 3. The description adds minimal parameter-level meaning by telling users to specify framework and scenarios, but the schema already documents framework, requirements, code, language, max_tokens, and temperature. No additional parameter semantics are provided beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a specific verb-object pair ('Generate test code') and clearly defines the deliverable as a 'complete, runnable test file' covering happy paths and edge cases. This distinguishes it from sibling tools like generate_docs and implement_code, making its purpose unambiguous.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives clear usage context: generate tests for a given implementation, and explicitly directs the agent to specify the testing framework and particular scenarios via the framework and requirements parameters. It lacks explicit exclusions or comparisons to alternatives, but the context is sufficient for an agent to know when to select it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Generate implementation code from a specification.
Use this tool when you need the local model to write a function, class, module, API endpoint, CRUD logic, glue code, or any other implementation artefact. Provide as much context and constraints as possible for the best results.
IMPORTANT: Always review the returned code before accepting it.
| Name | Required | Description | Default |
|---|---|---|---|
| context | No | Existing code, imports, type definitions, or file contents that the implementation must integrate with. Provide as much relevant context as possible for higher quality output. | |
| language | Yes | Programming language (e.g. 'python', 'typescript', 'rust'). | |
| max_tokens | No | Max tokens to generate. Defaults to server config. | |
| constraints | No | Coding conventions, performance requirements, library restrictions, edge cases to handle, or any other rules the implementation must follow. | |
| temperature | No | Sampling temperature override (0.0–1.0). Lower = more deterministic. | |
| task_description | Yes | Precise specification of what to implement. Include function signatures, expected behavior, input/output types, and edge cases. |
| Name | Required | Description |
|---|---|---|
| code | No | The generated artefact (code, tests, documentation). None on failure. |
| error | No | Human-readable error message. None on success. |
| success | Yes | Whether the generation completed successfully. |
| metadata | No | Auxiliary data: token usage, elapsed time, quality warnings, etc. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already signal readOnlyHint=true and openWorldHint=false, so the safety profile is covered. The description adds a useful behavioral cue with 'Always review the returned code before accepting it,' implying output may need validation. It does not richly describe side effects, output handling, or failure modes, but given the annotations, a mid score is appropriate.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact: a one-sentence purpose, a one-sentence usage condition, a one-sentence advice line, and a one-sentence warning. The structure front-loads the core action and keeps every sentence functional. It could be slightly tighter but is well-organized and not padded.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With a full input schema, output schema, and annotations covering safety, thedescription covers purpose, usage context, best-practice input guidance, and an important review warning. It does not mention pagination, side effects, or error cases, but those are less relevant for a code-generation tool and are partially covered by the output schema. The description is complete enough for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the input schema already documents all six parameters including context, constraints, task_description, language, temperature, and max_tokens. The description reinforces that context and constraints should be provided, but adds no new parameter-level meaning beyond the schema, so it earns the baseline score.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb ('Generate') and resource ('implementation code from a specification'), then enumerates concrete artifact types (function,class,module,API endpoint, CRUD logic, glue code), making it obvious what the tool does and how it differs from siblings like generate_tests, refactor_code, and fix_code. The purpose is precise and immediately actionable for an agent.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explicitly says 'Use this tool when you need the local model to write a function, class...', giving clear usage context. It advises providing context and constraints for best results. However, it does not state when not to use it or name alternative siblings explicitly, so it stops short of a fully explicit routing rule.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Apply a specific refactoring to existing code.
The local model will apply the requested transformation while preserving external API and behavior. Useful for mechanical refactors like renaming, extracting methods, converting patterns, etc.
IMPORTANT: Always review the returned code before accepting it.
| Name | Required | Description | Default |
|---|---|---|---|
| code | Yes | The existing code to refactor. | |
| language | Yes | Programming language of the code. | |
| max_tokens | No | Max tokens override. | |
| constraints | No | Rules or restrictions for the refactor. | |
| temperature | No | Temperature override. | |
| refactor_description | Yes | What refactoring to apply — e.g. 'extract method', 'rename variables to snake_case', 'convert class to dataclass', 'split into smaller functions'. |
| Name | Required | Description |
|---|---|---|
| code | No | The generated artefact (code, tests, documentation). None on failure. |
| error | No | Human-readable error message. None on success. |
| success | Yes | Whether the generation completed successfully. |
| metadata | No | Auxiliary data: token usage, elapsed time, quality warnings, etc. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already mark readOnlyHint=true, and the description adds useful behavioral context: the model preserves external API/behavior and the result must be reviewed before acceptance. The warning 'Always review the returned code before accepting it' communicates uncertainty quality without contradicting the read-only annotation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is compact and front-loaded: action first, then context, examples, and a critical safety warning. Every sentence contributes and there is no repetition of schema details.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The tool has a full input schema, output schema, and annotations; the description supplies the missing practical context—when to use it, what to preserve, and the need for review. Nothing essential for safe and correct invocation is absent.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the baseline is 3. The description adds value by giving concrete refactoring examples and framing refactor_description within preserving behavior, which helps an agent formulate meaningful transformation requests.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description opens with a specific verb and resource: 'Apply a specific refactoring to existing code.' It clarifies the scope with examples like renaming and extracting methods, and distinguishes from bug-fixing or implementation by emphasizing preservation of external API and behavior, though it does not name a sibling explicitly.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It states that the tool is useful for mechanical refactors, giving clear context for when it applies. However, it does not explicitly say when not to use it or name alternatives such as fix_code or implement_code, leaving exclusion to inference.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.
v1.0.0check_local_model_healthfix_codegenerate_docsgenerate_testsimplement_coderefactor_codeScored across 6 tools
Each tool targets a distinct workflow: implementation, test generation, refactoring, bug fixing, documentation, and health checking. Descriptions clearly separate concerns, so an agent should be able to select the right tool without confusion.
Tool names follow a consistent verb_noun snake_case pattern: implement_code, generate_tests, check_local_model_health, refactor_code, fix_code, generate_docs. The naming convention is uniform and predictable.
Six tools is a well-scoped set for a local-model code-assistance bridge. Each tool earns its place, covering the main code-generation and modification workflows without redundant or unnecessary additions.
The toolset covers the core lifecycle of generating, testing, refactoring, fixing, and documenting code, plus a health check. A minor gap is the lack of a general code review or explanation tool, but agents can work around it with the existing tools.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
Hosted MCP server connecting claude.ai, ChatGPT and other AI apps to your own computer
Remote MCP server for supportsheep: run AI interviews and manage support content for your blog.
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/shreyashp77/Llama-Bridge'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
