CoderSwap MCP Server
Enables ingestion and semantic search of research papers from arXiv, with automatic crawling, chunking, and embedding capabilities for building topic-specific knowledge bases
Supports ingestion of OpenAI research content and documentation into vector search projects for semantic search and knowledge base creation
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., "@CoderSwap MCP Server create a project for quantum computing research and ingest the latest papers from arXiv"
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.
CoderSwap MCP Server
Model Context Protocol (MCP) server that lets Claude (and any MCP-aware agent) stand up a topic-specific knowledge base end-to-end—project creation, ingestion, progress tracking, search validation, and lightweight session notes—without exposing low-level APIs.
Features
🚀 Create and list vector-search projects
📚 Ingest research summaries + URLs with auto-crawling, chunking, and embedding
🧠 Auto-ingest curated sources (crawl → chunk → embed) with relevance tuning handled by the CoderSwap platform team
🔍 Execute hybrid semantic search with intent-aware ranking
📊 Monitor ingestion jobs, capture blocked sources, and run quick search-quality spot checks
✨ Rich, formatted output optimized for AI agents
Installation
cd packages/mcp-server
npm install
npm run buildConfiguration
Set the following environment variables before launching the server:
CODERSWAP_BASE_URL(default:http://localhost:8000)CODERSWAP_API_KEY(required)DEBUG(optional: set totruefor detailed logging)
Running
Development (Local Backend)
# Set environment variables
export CODERSWAP_BASE_URL=http://localhost:8000
export CODERSWAP_API_KEY=cs_dev_nmVDJupuxflYYWd34HiRxbtxONul3hv1_f981
# Run the server
npm startClaude Desktop Configuration
Update your Claude Desktop config file:
macOS/Linux: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Local Development:
{
"mcpServers": {
"coderswap": {
"command": "node",
"args": ["C:/Users/tayav/CascadeProjects/CoderSwapIO/packages/mcp-server/dist/index.js"],
"env": {
"CODERSWAP_BASE_URL": "http://localhost:8000",
"CODERSWAP_API_KEY": "cs_dev_nmVDJupuxflYYWd34HiRxbtxONul3hv1_f981"
}
}
}
}Production:
{
"mcpServers": {
"coderswap": {
"command": "npx",
"args": ["-y", "@coderswap/mcp-server"],
"env": {
"CODERSWAP_BASE_URL": "https://api.coderswap.ai",
"CODERSWAP_API_KEY": "your_production_api_key"
}
}
}
}Available Tools
Project Management
coderswap_create_project– Create a new vector search projectcoderswap_list_projects– List accessible projects with document countscoderswap_get_project_stats– Pull basic stats (created_at, document totals)
Research & Ingestion
coderswap_research_ingest– Crawl, chunk, and embed vetted URLs (advanced tuning is managed by the platform team)coderswap_get_job_status– Poll ingestion job progress, crawl counts, blocked domains
Search & Validation
coderswap_search– Execute hybrid semantic search with ranked snippetscoderswap_test_search_quality– Run quick multi-query smoke tests (or a predefined suite) to gauge relevance
Session Continuity
coderswap_log_session_note– Record lightweight summaries (job_id, ingestion metrics, follow-ups) so humans stay in the loop
Guardrails & Security
The server loads
mcp_starter_prompt.yamlat startup and injects it as a non-removable system prompt.Startup fails if the prompt is missing, invalid, or tampered with (hash mismatch).
Advanced tuning endpoints are intentionally omitted; when deeper adjustments are required, Claude guides users to loop in the CoderSwap platform team.
All operations must go through the MCP tools; direct HTTP/DB access is disallowed.
Each tool:
✅ Validates inputs with Zod schemas
✅ Returns both structured data and AI-friendly text summaries
✅ Includes comprehensive error handling
✅ Logs operations for debugging (when DEBUG=true)
Example Usage
Autonomous Research Workflow
Claude can execute this workflow autonomously:
Create a project:
Use coderswap_create_project with name "AI Research"Ingest research content:
Use coderswap_research_ingest with URLs: - https://arxiv.org/abs/2103.00020 - https://openai.com/research/gpt-4Monitor progress (Claude keeps polling until complete):
Use coderswap_get_job_status to check ingestionSearch the knowledge base:
Use coderswap_search with query "transformer architecture"Optional: run a quick multi-query smoke test:
Use coderswap_test_search_quality with test queries or run_full_suite: trueLeave yourself a handoff note (e.g., sources blocked, next steps):
Use coderswap_log_session_note with project_id "proj_123", summary_text "Ingested 9/10 sources; FDA site blocked by robots.txt. Run follow-up after manual download." job_id "job_456" ingestion_metrics {"sources_succeeded": 9, "sources_failed": 1}
Output Format
Search results are formatted with rich details:
Found 5 result(s) for: "hybrid search"
🥇 Score: 85.2%
About hybrid search | Vertex AI
Vector Search supports hybrid search...
🥈 Score: 72.1%
Hybrid Search | Weaviate
Hybrid search combines semantic and keyword...
🥉 Score: 68.4%
...Debugging
Enable debug logging:
export DEBUG=true
npm startLogs are written to stderr and include:
Timestamps
Operation details
Error messages with context
Development
# Install dependencies
npm install
# Build TypeScript
npm run build
# Watch mode (for development)
npm run devArchitecture
Claude Desktop → MCP Server (stdio) → CoderSwap Backend API → Oracle ADW 23ai
↓
- Tool validation (Zod)
- Error handling
- Response formattingWith the MCP server, Claude can autonomously build, test, and optimize vector knowledge bases in minutes! 🚀
Available Tools
8 toolscoderswap_create_projectCreate CoderSwap ProjectB
Create a new vector search project in CoderSwap
Inputs: name (required), description. Returns: name, status, project_id
coderswap_get_job_statusGet CoderSwap Job StatusB
Check the status of a research ingestion job
Inputs: job_id (required). Returns: state, job_id, failed_count, crawled_count
coderswap_get_project_statsGet CoderSwap Project StatsC
Get statistics and information about a specific project
Inputs: project_id (required). Returns: name, doc_count, created_at, project_id
coderswap_list_projectsList CoderSwap ProjectsB
List all CoderSwap projects available to your API key
Inputs: no named parameters. Returns: count, projects
coderswap_log_session_noteLog Session NoteC
Record lightweight ingestion summary for session continuity (non-DSL)
Inputs: project_id (required), summary_text (required), job_id, ingestion_metrics, +1 more. Returns: note_id, timestamp, project_id
coderswap_research_ingestCoderSwap Research IngestC
Submit research summary and URLs for web crawling, chunking, embedding, and optional DSL generation
Inputs: project_id (required), urls (required), research_summary, intent, +2 more. Returns: job_id, status, project_id
coderswap_searchCoderSwap Hybrid SearchC
Execute a hybrid search query against a CoderSwap project using DSL-powered ranking
Inputs: project_id (required), query (required), top_k, snippet_length. Returns: query, results, result_count
coderswap_validate_searchValidate CoderSwap Search QualityB
Run validation queries to test search quality and coverage (non-DSL quality check)
Inputs: project_id (required), test_queries, run_full_suite. Returns: queries_tested, average_top_score, zero_result_queries
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 8 tools
Each tool has a clearly distinct purpose with no ambiguity. The tools cover different aspects of the workflow: project management (create, list, get stats), job monitoring (get job status), ingestion (research_ingest, log_session_note), and search operations (search, validate_search). There is no overlap in functionality that would cause misselection.
All tools follow a consistent 'coderswap_verb_noun' pattern throughout, using snake_case uniformly. The verbs are descriptive and appropriate for each action (e.g., create_project, get_job_status, list_projects), making the naming highly predictable and readable.
With 8 tools, the count is well-scoped for a vector search project management server. Each tool earns its place by covering essential operations: project lifecycle (create, list, get stats), ingestion workflows (research_ingest, log_session_note), job monitoring, and search functionality (search, validate). This is neither too sparse nor bloated.
The tool set provides comprehensive coverage for core CoderSwap workflows, including project CRUD, ingestion, and search. A minor gap exists in the lack of update or delete operations for projects, which might limit full lifecycle management, but agents can likely work around this given the server's focus on ingestion and search.
Related MCP Connectors
Shared knowledge base for AI agents. Semantic search across agents, no setup required — just a URL.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
Curated knowledge API for AI agents - skill packs, semantic search, validated patterns.
Shared project context for AI agents and teams: docs, tasks, and messages that stay current.