Screens public GitHub repositories and pull requests, fetching metadata and files for local risk analysis.
Click on "Deploy 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., "@gitlumen-mcp Screen https://github.com/facebook/react for risks"
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.
GitLumen MCP Server is a Node.js project that exposes a GitLumen-style review intelligence layer through the Model Context Protocol (MCP), so AI agents can call it as tools.
This project focuses on:
AI Agent / MCP Client
-> GitLumen MCP Server
-> GitHub public repo / PR reader
-> local risk analyzer
-> GitLumen-style reportThis project intentionally does not execute onchain transactions yet and does not use Base MCP send_calls. A Base MCP custom plugin can be attached in Path 2 after this intelligence server is ready.
MCP stdio server that can be used by Claude Desktop, Cursor, Claude Code, or other MCP clients.
Screens public GitHub repository URLs.
Screens GitHub Pull Request URLs /pull/<number>.
No GitHub token required for small/medium public repositories.
Optional GITHUB_TOKEN for higher rate limits and private repositories (depending on token scope).
Local analyzer: source code is not sent to external LLMs.
Produces:
risk score
category risk map
findings
review chapters
decision questions
merge-readiness signal
recommended next actions
Stores reports locally in .gitlumen-mcp/reports/*.json.
Includes a CLI for local testing without an MCP client.
Related MCP server: agentic-sdlc-mcp
gitlumen-mcp-server/
|- package.json
|- README.md
|- .env.example
|- examples/
| |- claude_desktop_config.example.json
| \- cursor_mcp.example.json
|- docs/
| |- ARCHITECTURE.md
| \- TOOLS.md
\- src/
|- index.js # MCP stdio server entrypoint
|- cli.js # CLI local test
|- doctor.js # environment checker
|- config.js
|- types.js
|- services/
| |- github.js # GitHub API + raw file loader
| |- analyzer.js # local heuristic risk engine
| |- gitlumen.js # service orchestrator
| \- reportStore.js # local report persistence
\- utils/
|- githubUrl.js
|- ids.js
\- text.jsNode.js 20+
npm
Internet access to fetch metadata/files from GitHub
Check Node version:
node -vIf your version is Node 18 or below, upgrade to Node 20+.
Open the project directory:
cd gitlumen-mcp-serverInstall dependencies:
npm installCopy env example:
cp .env.example .envFill optional values:
GITHUB_TOKEN=ghp_xxx_or_fine_grained_token
GITLUMEN_MCP_DATA_DIR=.gitlumen-mcp
GITLUMEN_MAX_FILE_BYTES=120000For public repositories, GITHUB_TOKEN can be empty. A token is still recommended to avoid low GitHub rate limits.
npm run doctorExpected output:
GitLumen MCP Doctor
✅ Node version: v20.x.x
✅ GITHUB_TOKEN configured: no (public unauthenticated mode)
✅ Data directory: /path/to/gitlumen-mcp-server/.gitlumen-mcp
✅ Reports directory writable: /path/to/gitlumen-mcp-server/.gitlumen-mcp/reportsnpm run sampleThis command generates a report from a local fixture so you can verify analyzer and report-store behavior without GitHub connectivity.
npm run screen -- https://github.com/modelcontextprotocol/typescript-sdk quicknpm run screen -- https://github.com/modelcontextprotocol/typescript-sdk/pull/1 quickquick = fastest, fewer files
standard = balanced defaultExamples:
npm run screen -- https://github.com/owner/repo standard
npm run screen -- https://github.com/owner/repo quick mainAfter completion, CLI prints a markdown report and saves JSON to:
.gitlumen-mcp/reports/<reportId>.jsonnpm run list -- 10Take a reportId, then:
npm run report -- glr_xxxxxxxxxxxxxxxx markdownOr full JSON:
npm run report -- glr_xxxxxxxxxxxxxxxx jsonThe MCP server uses stdio, so it is normally started by an MCP client instead of being run manually.
node /ABSOLUTE/PATH/TO/gitlumen-mcp-server/src/index.jsTo debug MCP protocol, use MCP Inspector:
npm run inspectThen open the Inspector URL printed in terminal.
This project also includes a Streamable HTTP transport endpoint at /mcp.
Run locally:
npm run start:httpEnvironment variables:
PORT=3333
HOST=0.0.0.0
MCP_AUTH_TOKEN=replace_with_a_long_random_tokenMCP_AUTH_TOKEN is optional but strongly recommended for production.
When set, clients must send Authorization: Bearer <token>.
Health check:
curl -s http://localhost:3333/healthProduction deployment guide:
PM2 example config: examples/ecosystem.pm2.example.cjs
Client configuration templates (Copilot / VS Code / Codex):
VS Code remote endpoint template: examples/vscode_mcp.gitlumen.remote.example.json (copy into .vscode/mcp.json, which is gitignored)
Open Claude Desktop config.
Common location:
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.jsonAdd:
{
"mcpServers": {
"gitlumen": {
"command": "node",
"args": ["/ABSOLUTE/PATH/TO/gitlumen-mcp-server/src/index.js"],
"env": {
"GITHUB_TOKEN": "optional_github_token_here",
"GITLUMEN_MCP_DATA_DIR": "/ABSOLUTE/PATH/TO/gitlumen-mcp-server/.gitlumen-mcp"
}
}
}
}Replace /ABSOLUTE/PATH/TO/... with your real path.
Restart Claude Desktop.
Example prompt:
Use GitLumen to screen https://github.com/modelcontextprotocol/typescript-sdk with quick scope. Return the risk map and top findings.Create or edit Cursor MCP config (format may vary by Cursor version):
{
"mcpServers": {
"gitlumen": {
"command": "node",
"args": ["/ABSOLUTE/PATH/TO/gitlumen-mcp-server/src/index.js"],
"env": {
"GITHUB_TOKEN": "optional_github_token_here"
}
}
}
}Restart Cursor, then ask the agent to use GitLumen tools.
screen_repositoryScreen a repository or PR.
Input:
{
"repoUrl": "https://github.com/owner/repo",
"scope": "standard",
"output": "compact"
}For PR:
{
"repoUrl": "https://github.com/owner/repo/pull/123",
"scope": "quick",
"output": "markdown"
}Output modes:
compact = concise JSON for agent replies
markdown = full markdown report
json = full JSON reportget_review_reportFetch a previous report by reportId.
{
"reportId": "glr_xxxxxxxxxxxxxxxx",
"output": "markdown"
}list_review_reportsList local reports.
{
"limit": 20
}get_repository_structureGet repository/PR structure without generating a full risk report.
{
"repoUrl": "https://github.com/owner/repo",
"limit": 300
}explain_gitlumen_mcp_flowExplain Path 1 flow and how Path 2 Base MCP can be attached later.
The local analyzer reads:
repository metadata
default branch
recursive tree
selected source/config files
PR metadata and changed files (for PR URLs)
Then it generates signals:
language/framework detection
dependency surface
lockfile presence
lifecycle script risk
test presence
CI presence
Dockerfile/container risk
possible hardcoded secret patterns
dynamic code execution
command execution pattern
SQL interpolation pattern
GitHub Actions supply-chain pattern
merge-readiness estimate
Risk categories:
security
dependencies
tests
architecture
operations
maintainabilitySeverity:
critical
high
medium
low
info{
"reportId": "glr_abc123...",
"risk": {
"score": 42,
"level": "medium",
"mergeReadiness": "review_required",
"categoryScores": {
"security": 24,
"dependencies": 13,
"tests": 24,
"architecture": 0,
"operations": 13,
"maintainability": 5
}
},
"summary": "The repository/PR has medium risk signals...",
"findings": [],
"decisionQuestions": [],
"recommendations": []
}Repo/PR intelligence
Risk map
Review chapters
Decision questions
Report retrievalBase MCP get_wallets
GitLumen quote endpoint
GitLumen prepare endpoint
Base MCP send_calls
Review credit purchase
Reward claim
Reviewer reputationThis project is intentionally standalone for Path 1 first. Later, Path 2 can read reportId and connect it with onchain payment/reward/reputation flows.
Unable to reach GitHub API or fetch failedCheck internet connection, DNS, proxy/VPN, or retry. For offline verification:
npm run sampleGitHub API 403 rate limit exceededAdd GITHUB_TOKEN in .env or MCP client config.
Only github.com repositories are supportedThis prototype does not support GitLab/Bitbucket yet. Add a new adapter in src/services/github.js or create a separate service.
Check:
args path is absolute.
npm install has been run.
Node 20+ is installed.
MCP client was restarted.
Verify with npm run inspect.
Run:
npm run doctorEnsure .gitlumen-mcp/reports is writable.
Edit:
src/services/analyzer.jsEdit:
src/services/github.jsEdit:
src/services/gitlumen.jsPotential production direction:
screen_repository MCP tool
-> GitLumen hosted API /v1/screenings
-> GitLumen Review Intelligence Engine
-> reportId
-> get_review_report MCP toolDo not commit .env.
Do not hardcode GitHub tokens in publicly shared config.
For private repositories, use least-privilege fine-grained GitHub tokens.
Local reports may contain paths, findings, and snippet metadata. Store them securely for private repositories.
MIT
Explain how this Path 1 MCP server fits into GitLumen and how it later connects to Base MCP Path 2.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full behavioral burden. 'Explain' clearly signals an informational, non-mutating action with no side effects, which is the core behavioral trait an agent needs. It does not specify whether the explanation is static or dynamically generated, but for a zero-parameter explanatory tool this is a minor omission, not a dangerous gap.
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 a single sentence that front-loads the verb and clearly states both parts of the scope (fit into GitLumen, later connection to Base MCP Path 2). It contains no fluff and does not repeat information already available in the name or schema.
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, no-output-schema tool that exists solely to explain a concept, this description is essentially complete: it tells the agent what the tool explains and how the explanation is framed. A note about the form of the output (e.g., prose explanation) would be slightly helpful but is not necessary 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?
The input schema is empty and there are zero parameters, so there is nothing for the description to add over the schema. The baseline of 4 is appropriate because no parameter documentation is needed or expected.
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 uses a specific action ('Explain') and identifies exactly what is explained: how this Path 1 MCP server fits into GitLumen and connects to Base MCP Path 2. It is clearly distinct from sibling tools like get_repository_structure and get_review_report, which focus on data retrieval rather than conceptual explanation. It does not explicitly name a sibling, but the resource and action make the 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?
There is no explicit when-to-use statement, exclusions, or alternatives. The imperative 'Explain...' implies the tool is appropriate when a user asks about the GitLumen MCP architecture or Path 2 connection, but the decision is left to inference. Since the sibling tools are not plausible alternatives for an explanation request, the absence of alternation guidance is not severe, but the guidance is still mostly implied.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Fetch public GitHub repository or PR structure without generating a full risk report.
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No | Maximum tree entries to return. | |
| branch | No | Optional branch/ref. | |
| repoUrl | Yes | GitHub repository or pull request URL. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the burden of behavioral disclosure. It indicates a read-only public fetch and clarifies that no risk report is generated, but it omits details about rate limits, truncation due to the 'limit' parameter, or error behavior for invalid/private URLs.
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 a single efficient sentence with no filler. The core action and the distinguishing negative clause are front-loaded, and every word earns its place.
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?
No output schema exists, and the description does not explain what the returned 'structure' looks like, how truncation works with 'limit', or what happens on errors. The complete parameter schema helps, but the absence of return-value or edge-case information leaves the definition incomplete for an agent.
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 only that the repository is 'public', which slightly qualifies repoUrl, but it does not enhance understanding of 'limit' or 'branch' beyond the schema's existing parameter descriptions.
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 uses a specific verb, 'Fetch', and names the resource, 'public GitHub repository or PR structure', making the action clear. The clause 'without generating a full risk report' explicitly distinguishes it from sibling report-generation 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 implies usage when only repository/PR structure is needed and explicitly says it does not generate a full risk report. However, it does not name alternative tools or provide explicit when-not-to-use guidance, so it stops 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.
Get a previously generated GitLumen MCP report by reportId.
| Name | Required | Description | Default |
|---|---|---|---|
| output | No | compact | |
| reportId | Yes | Report id returned by screen_repository, for example glr_abcd1234abcd1234 |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full behavioral burden. It clearly signals a read-style retrieval of an existing report and rules out generation, but it does not disclose failure behavior for missing/invalid reportIds or describe the response format. For a simple getter, this is adequate but not rich.
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 a single sentence with no filler. It front-loads the action, resource, and key parameter, making every word useful.
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 straightforward get-by-id tool, the description plus schema cover the core information needed to select and invoke it: what it retrieves, the required ID, and the output format options. It only lacks explicit guidance about using list_review_reports to discover report IDs, which is a minor gap.
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?
The schema already documents reportId well, including an example and its source, and the output parameter is self-describing through its enum and default. The description itself adds little beyond restating 'by reportId' and does not explain the meaning or trade-offs of the output options.
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 ('Get') and a clear resource ('previously generated GitLumen MCP report'), keyed by reportId. The phrase 'previously generated' distinguishes it from generation-focused siblings like screen_repository without requiring the agent to inspect schemas.
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 implies this should be used after a report has been generated, since reportId comes from screen_repository. However, it does not explicitly mention sibling alternatives like list_review_reports for discovering report IDs, nor does it state when this tool should not be used.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List previously generated GitLumen MCP reports stored locally.
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description is the only behavioral signal. It does disclose that the operation is local and that it only surfaces previously generated reports, implying no new generation or remote scanning. It does not, however, describe result ordering, empty-state behavior, or whether the operation is purely read-only, which would be helpful without annotations.
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 a single, tight sentence with no filler. Every word contributes meaning: 'previously generated,' 'stored locally,' and 'reports' all constrain the tool's behavior.
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 simple list tool with one optional parameter and no output schema, the description covers the essential context: what is listed, where it comes from, and that it is limited to prior local report generation. Additional return-format details would be nice but are not critical 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?
The schema has one optional 'limit' parameter with default, minimum, and maximum, but no additional description coverage. The parameter name plus constraints make its meaning reasonably clear, so the description does not need to restate it. Still, the description adds no value about how 'limit' affects the returned list.
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 uses a specific verb ('List') and names a precise resource ('previously generated GitLumen MCP reports stored locally'). This clearly separates it from siblings like get_review_report (fetching one report) and screen_repository (scanning a repository), so an agent can identify its role immediately.
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 wording implies the tool is for enumerating already-generated local reports, which hints that it would be used before get_review_report. However, it does not explicitly state when to prefer this over alternatives or mention any exclusions, leaving some routing to inference.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Screen a public GitHub repository or GitHub pull request URL and generate a GitLumen-style risk report. Supports repo URLs and /pull/ URLs.
| Name | Required | Description | Default |
|---|---|---|---|
| scope | No | Screening depth. quick downloads fewer files; standard downloads more files. | standard |
| branch | No | Optional branch/ref. Ignored for PR URLs unless GitHub needs fallback. | |
| output | No | Response format returned to the MCP client. | compact |
| repoUrl | Yes | GitHub repository URL, for example https://github.com/owner/repo or https://github.com/owner/repo/pull/123 | |
| maxFiles | No | Optional hard cap for files downloaded and scanned. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description must carry full behavioral disclosure. It states the action and supported URL shapes, but does not disclose that the tool downloads/scans file contents, the read-only nature beyond 'screen', network/resource implications, or any constraints. This is a meaningful gap for a tool with side effects like file downloads.
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?
Two sentences, front-loaded with the core purpose, followed by supported URL formats. No filler; all sentences add value and the description is appropriately sized for a tool with a detailed schema.
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 combination of description and schema is enough to invoke the tool, but the description alone does not clarify how this relates to sibling tools, what a 'GitLumen-style risk report' contains, or the behavior/impact of downloads. With no output schema, a bit more detail about report output would improve completeness.
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 all five parameters (repoUrl, scope, branch, output, maxFiles) are already documented with types, defaults, enums, and a repository URL example. The description adds no parameter-level details, which is acceptable given the schema's rich coverage.
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 uses a specific verb ('Screen'), names the exact resource ('public GitHub repository or GitHub pull request URL'), and states the output ('GitLumen-style risk report'). This clearly differentiates it from siblings like get_repository_structure and list_review_reports.
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 implies when to use this tool (when a risk report on a public repo/PR is needed) but gives no explicit when-not-to-use guidance or alternatives. It does not mention that get_review_report/list_review_reports serve existing reports or that get_repository_structure covers structure-only requests. The 'public' qualifier is the only condition stated.
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.
v1.0.0explain_gitlumen_mcp_flowget_repository_structureget_review_reportlist_review_reportsscreen_repositoryScored across 5 tools
Each tool has a clearly distinct purpose: generating reports, retrieving specific reports, listing reports, fetching repository structure, and explaining the flow. No overlap.
All tool names follow a consistent verb_noun pattern in snake_case (e.g., screen_repository, get_review_report, list_review_reports), making them predictable and easy to distinguish.
With 5 tools, the server is well-scoped for its purpose. Each tool serves a necessary function without bloat or insufficiency.
Core operations are covered: generate, get, list, and structure exploration. Missing delete or update functionality for reports, but the server's focus on one-time generation and review makes this a minor gap.
Security reviews for coding agents: diffs checked against your org policy and live infrastructure.
Pay-per-call cybersecurity for AI agents: vuln scans, threat intel, compliance, code security.
Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language.
Scan GitHub-hosted AI skills for vulnerabilities: prompt injection, malware, OWASP LLM Top 10.
